Added option to output an empty image instead of sending an exception if the image generation by DALLe3 fails (probably due to a content filter)

README.md

@@ -1,5 +1,10 @@
 # ComfyUI-NegiTools
 
+> [!IMPORTANT]
+> "Depth Estimation by Marigold (experimental)" module is not maintained and will be discontinued in the future; 
+> if you would like to continue using Marigold, please consider using this alternative choice.
+> https://github.com/kijai/ComfyUI-Marigold
+
 ## Installation
 
 - Install dependencies: pip install -r requirements.txt

__init__.py

@@ -9,6 +9,10 @@ from .negi.open_pose_to_point_list import OpenPoseToPointList
 from .negi.point_list_to_mask import PointListToMask
 from .negi.depth_estimation_by_marigold import DepthEstimationByMarigold
 from .negi.stereo_image_generator import StereoImageGenerator
+from .negi.image_reader_writer import RandomImageLoader, SaveImageToDirectory
+from .negi.detect_face_rotation_for_inpainting import DetectFaceRotationForInpainting
+from .negi.openai_gpt4v import OpenAiGpt4v
+from .negi.openai_gpt import OpenAiGpt
 
 NODE_CLASS_MAPPINGS = {
     "NegiTools_OpenAiDalle3": OpenAiDalle3,
@@ -23,6 +27,11 @@ NODE_CLASS_MAPPINGS = {
     "NegiTools_PointListToMask": PointListToMask,
     "NegiTools_DepthEstimationByMarigold": DepthEstimationByMarigold,
     "NegiTools_StereoImageGenerator": StereoImageGenerator,
+    "NegiTools_RandomImageLoader": RandomImageLoader,
+    "NegiTools_SaveImageToDirectory": SaveImageToDirectory,
+    "NegiTools_DetectFaceRotationForInpainting": DetectFaceRotationForInpainting,
+    "NegiTools_OpenAiGpt4v": OpenAiGpt4v,
+    "NegiTools_OpenAiGpt": OpenAiGpt,
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
@@ -38,4 +47,9 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "NegiTools_PointListToMask": "Point List to Mask 🧅",
     "NegiTools_DepthEstimationByMarigold": "Depth Estimation by Marigold (experimental) 🧅",
     "NegiTools_StereoImageGenerator": "Stereo Image Generator 🧅",
+    "NegiTools_RandomImageLoader": "Random Image Loader 🧅",
+    "NegiTools_SaveImageToDirectory": "Save Image to Directory 🧅",
+    "NegiTools_DetectFaceRotationForInpainting": "Detect Face Rotation for Inpainting 🧅",
+    "NegiTools_OpenAiGpt4v": "OpenAI GPT4V 🧅",
+    "NegiTools_OpenAiGpt": "OpenAI GPT 🧅",
 }

negi/depth_estimation_by_marigold.py

@@ -1,6 +1,7 @@
 import os
 import sys
 import subprocess
+import gc
 
 import numpy as np
 import torch
@@ -76,6 +77,10 @@ class DepthEstimationByMarigold:
                    "bare (recommended)",
                    "venv (if \"bare\" doesn't work)",
                 ],),
+                "depth_exponent": ("FLOAT", {
+                    "default": 1.0, "min": 0.1, "max": 3.0, "step": 0.1, "round": 0.01, "display": "slider"
+                }),
+                "invert": ("BOOLEAN", {"default": False}),
             }
         }
 
@@ -85,7 +90,7 @@ class DepthEstimationByMarigold:
     OUTPUT_NODE = False
     CATEGORY = "Generator"
 
-    def doit(self, image, infer_passes, denoise_steps, seed, runtime):
+    def doit(self, image, infer_passes, denoise_steps, seed, runtime, depth_exponent, invert):
         use_venv = runtime.startswith("venv")
         self.__check_environment(use_venv)
 
@@ -110,6 +115,9 @@ class DepthEstimationByMarigold:
                 "output_dir_name": os.path.abspath(output_dir)
             })
 
+        # TODO It seems to be very bad idea, but it works; Try in-process execution
+        gc.collect()
+        torch.cuda.empty_cache()
         subprocess.run(["bash", os.path.join("work", "infer.sh")], cwd=self.rep_dir)
 
         im1 = np.load(os.path.join(output_dir, "depth_npy", "image_pred.npy")).astype(np.float32)
@@ -118,5 +126,8 @@ class DepthEstimationByMarigold:
         if im1_min == im1_max:
             im1_max = im1_min + 1.0
         im1 = (im1 - im1_min) * (1.0 / (im1_max - im1_min))
+        im1 = np.power(im1, depth_exponent)
+        if invert:
+            im1 = 1.0 - im1
 
         return (torch.from_numpy(np.expand_dims(np.stack([im1, im1, im1], axis=-1), axis=0)),)

negi/detect_face_rotation_for_inpainting.py

@@ -0,0 +1,118 @@
+import json
+
+import numpy as np
+import torch
+
+
+class DetectFaceRotationForInpainting:
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "parts": ("STRING", {"multiline": False, "default": ""}),
+                "image": ("IMAGE",),
+                "radius_scale": ("FLOAT", {
+                    "default": 1.0,
+                    "min": 0.1,
+                    "max": 5.0,
+                    "step": 0.01,
+                    "round": 0.001,
+                    "display": "number"
+                }),
+                "overwrite_rotation": (["None", "0", "90", "180", "270"],),
+            }
+        }
+
+    RETURN_TYPES = ("INT", "MASK", "INT")
+    RETURN_NAMES = ("ROTATION_INV", "MASK", "ROTATION")
+    FUNCTION = "doit"
+    OUTPUT_NODE = False
+    CATEGORY = "utils"
+
+    @staticmethod
+    def get_face(xw, yw, radius_scale, parts):
+        x = 0.0
+        y = 0.0
+        n = 0
+        radius = 0
+        rot = 0
+
+        for name in ["Nose", "REye", "LEye", "REar", "LEar"]:
+            if name in parts:
+                x += parts[name]["x"]
+                y += parts[name]["y"]
+                n += 1
+
+        if n != 0:
+            x = x / n
+            y = y / n
+            for name in ["Nose", "REye", "LEye", "REar", "LEar"]:
+                if name in parts:
+                    x0 = x * xw
+                    y0 = y * yw
+                    x1 = parts[name]["x"] * xw
+                    y1 = parts[name]["y"] * yw
+                    radius = max(radius, int(np.sqrt((x1 - x0) * (x1 - x0) + (y1 - y0) * (y1 - y0)) * radius_scale))
+
+        if n != 0 and "Neck" in parts:
+            x0 = x * xw
+            y0 = y * yw
+            x1 = parts["Neck"]["x"] * xw
+            y1 = parts["Neck"]["y"] * yw
+            if abs(x1 - x0) < abs(y1 - y0):
+                rot = 0 if y0 < y1 else 180
+            else:
+                rot = 90 if x0 < x1 else 270
+
+        return x, y, radius, rot
+
+    @staticmethod
+    def rotate(rot, x, y):
+        if rot == 0:
+            return x, y
+        if rot == 90:
+            return y, 1 - x
+        if rot == 180:
+            return 1 - x, 1 - y
+        if rot == 270:
+            return 1 - y, x
+
+    def doit(self, parts, image, radius_scale, overwrite_rotation):
+        parts_list = json.loads(parts)
+        xw = image.shape[2]
+        yw = image.shape[1]
+
+        x = 0.0
+        y = 0.0
+        radius = 0
+        rot = 0
+        for parts in parts_list:
+            t_x, t_y, t_radius, t_rot = self.get_face(xw, yw, radius_scale, parts)
+            if t_radius > radius:
+                x = t_x
+                y = t_y
+                radius = t_radius
+                rot = t_rot
+
+        if overwrite_rotation != "None":
+            rot = int(overwrite_rotation)
+
+        rot_inv = (0 if rot == 0 else 360 - rot)
+        x_r, y_r = self.rotate(rot_inv, x, y)
+
+        xw_r = (xw if rot == 0 or rot == 180 else yw)
+        yw_r = (yw if rot == 0 or rot == 180 else xw)
+
+        if radius == 0:
+            return rot_inv, torch.from_numpy(np.zeros((1, yw_r, xw_r), dtype=np.float32)), rot
+
+        px = (np.reshape(np.arange(xw_r, dtype=np.float32), (1, -1))
+              * np.ones((yw_r, 1), dtype=np.float32))
+        py = (np.reshape(np.arange(yw_r, dtype=np.float32), (-1, 1))
+              * np.ones((1, xw_r), dtype=np.float32))
+        d2 = np.power(px - x_r * xw_r, 2.0) + np.power(py - y_r * yw_r, 2.0)
+        mask = torch.from_numpy(np.reshape(d2 <= radius * radius, (1, yw_r, xw_r)).astype(np.float32))
+        return rot_inv, mask, rot

negi/image_reader_writer.py

@@ -0,0 +1,83 @@
+import glob
+import os
+
+import torch
+from PIL import Image
+import torchvision
+from torchvision.transforms import functional as TF
+
+from . import utils
+
+
+class RandomImageLoader:
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "directory": ("STRING", {"multiline": False, "default": "./input"}),
+                "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "doit"
+    OUTPUT_NODE = False
+    CATEGORY = "utils"
+
+    def doit(self, directory, seed):
+        directory = utils.get_directory(directory)
+        print("RandomImageLoader: directory = %s" % directory)
+
+        files = (glob.glob(os.path.join(directory, "*.png")) +
+                 glob.glob(os.path.join(directory, "*.jpg")) +
+                 glob.glob(os.path.join(directory, "*.jpeg")))
+
+        if len(files) == 0:
+            raise ValueError("Specified directory does not contain any image files")
+
+        file = files[seed % len(files)]
+        print("RandomImageLoader: load %s; in %d files" % (file, len(files)))
+
+        im0 = Image.open(file)
+        im1 = TF.to_tensor(im0.convert("RGBA"))
+        im1[:3, im1[3, :, :] == 0] = 0
+
+        images = torch.stack([im1])
+        images = images.permute(0, 2, 3, 1)
+        images = images[:, :, :, :3]
+        return (images,)
+
+
+class SaveImageToDirectory:
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "directory": ("STRING", {"multiline": False, "default": "./output"}),
+                "image": ("IMAGE",),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    FUNCTION = "doit"
+    OUTPUT_NODE = True
+    CATEGORY = "utils"
+
+    def doit(self, directory, image):
+        directory = utils.get_directory(directory)
+        print("SaveImageToDirectory: directory = %s" % directory)
+
+        next_index = utils.find_next_index(directory)
+        file_name = os.path.join(directory, "out.%06d.png" % next_index)
+        print("SaveImageToDirectory: save to %s" % file_name)
+
+        im0 = torchvision.transforms.functional.to_pil_image(torch.permute(image[0], (2, 0, 1)))
+        im0.save(file_name)
+
+        return (image,)

negi/open_pose_to_point_list.py

@@ -1,8 +1,11 @@
 import json
 
 import numpy as np
-from controlnet_aux import OpenposeDetector
-from controlnet_aux.util import HWC3, resize_image
+import cv2
+import torch
+from .repos.controlnet_aux.src.controlnet_aux import OpenposeDetector
+from .repos.controlnet_aux.src.controlnet_aux.util import HWC3
+from .repos.controlnet_aux.src.controlnet_aux.open_pose import draw_poses
 
 
 _names = [
@@ -17,6 +20,19 @@ _names = [
 _name_to_index = {name: i for i, name in enumerate(_names)}
 
 
+def _resize_image(input_image, resolution):
+    H, W, C = input_image.shape
+    H = float(H)
+    W = float(W)
+    k = float(resolution) / min(H, W)
+    H *= k
+    W *= k
+    H = int(np.round(H / 64.0)) * 64
+    W = int(np.round(W / 64.0)) * 64
+    img = cv2.resize(input_image, (W, H), interpolation=cv2.INTER_LANCZOS4 if k > 1 else cv2.INTER_AREA)
+    return img, H, W
+
+
 class OpenPoseToPointList:
     def __init__(self):
         self.open_pose = OpenposeDetector.from_pretrained("lllyasviel/Annotators")
@@ -35,8 +51,8 @@ class OpenPoseToPointList:
             },
         }
 
-    RETURN_TYPES = ("STRING",)
-    RETURN_NAMES = ("POINT_LIST",)
+    RETURN_TYPES = ("STRING", "IMAGE")
+    RETURN_NAMES = ("POINT_LIST", "IMAGE")
     FUNCTION = "doit"
     OUTPUT_NODE = False
     CATEGORY = "utils"
@@ -44,9 +60,11 @@ class OpenPoseToPointList:
     def doit(self, image, detect_resolution, method):
         input_image = (np.fmax(0.0, np.fmin(1.0, image.to('cpu').detach().numpy()[0])) * 255.0).astype(np.uint8)
         input_image = HWC3(input_image)
-        input_image = resize_image(input_image, detect_resolution)
+        input_image, H, W = _resize_image(input_image, detect_resolution)
 
         poses = self.open_pose.detect_poses(input_image, include_hand=False, include_face=False)
+        img = draw_poses(poses, H, W, draw_hand=False, draw_face=False)
+        img = torch.from_numpy(np.expand_dims(HWC3(img) * (1.0 / 255), axis=0))
 
         if method == "face":
             ret = []
@@ -83,4 +101,4 @@ class OpenPoseToPointList:
         else:
             raise ValueError()
 
-        return (json.dumps(ret, indent=2),)
+        return (json.dumps(ret, indent=2), img)

negi/openai_dalle3.py

@@ -1,11 +1,15 @@
 import openai
 import base64
 import io
+import os
+import json
 
 import torch
 from PIL import Image
 from torchvision.transforms import functional as TF
 
+from . import utils
+
 
 class OpenAiDalle3:
     def __init__(self):
@@ -25,7 +29,13 @@ class OpenAiDalle3:
                 "prompt": ("STRING", {
                     "multiline": True,
                     "default": "great picture"
-                })
+                }),
+                "quality": (["HD", "Standard"],),
+                "style": (["vivid", "natural"],),
+                "retry": ("INT", {"default": 0, "min": 0, "max": 5}),
+                "auto_save": ("BOOLEAN", {"default": False}),
+                "auto_save_dir": ("STRING", {"multiline": False, "default": "./output_dalle3"}),
+                "blank_image_if_failed": ("BOOLEAN", {"default": False}),
             }
         }
 
@@ -35,19 +45,53 @@ class OpenAiDalle3:
     OUTPUT_NODE = True
     CATEGORY = "Generator"
 
-    def doit(self, resolution, dummy_seed, prompt):
+    def doit(
+            self, resolution, dummy_seed, prompt, quality, style, retry, auto_save, auto_save_dir,
+            blank_image_if_failed):
+
         if (self.__cache_image is None or
                 self.__previous_resolution != resolution or self.__previous_seed != dummy_seed or
                 self.__previous_prompt != prompt):
-            r0 = self.__client.images.generate(
-                model="dall-e-3",
-                prompt=prompt,
-                size=resolution,
-                quality="hd",  # "standard"
-                n=1,
-                response_format="b64_json"
-            )
+            r0 = None
+            for retry_count in range(retry + 1):
+                try:
+                    r0 = self.__client.images.generate(
+                        model="dall-e-3",
+                        prompt=prompt,
+                        size=resolution,
+                        quality="hd" if quality == "HD" else "standard",
+                        style="vivid" if style == "vivid" else "natural",
+                        n=1,
+                        response_format="b64_json"
+                    )
+                    break
+                except openai.BadRequestError as ex:
+                    if retry_count >= retry:
+                        print("OpenAiDalle3: received BadRequestError: %s" % (json.dumps(ex.response.json()),))
+                        if blank_image_if_failed:
+                            widths = resolution.split("x")
+                            xw, yw = int(widths[0]), int(widths[1])
+                            return torch.zeros(1, yw, xw, 4), xw, yw, "(BadRequestError)"
+                        raise ex
+                    print("OpenAiDalle3: received BadRequestError, retrying... #%d : %s" % (
+                        retry_count + 1, json.dumps(ex.response.json())))
+
             im0 = Image.open(io.BytesIO(base64.b64decode(r0.data[0].b64_json)))
+
+            if auto_save:
+                directory = utils.get_directory(auto_save_dir)
+                next_index = utils.find_next_index(directory)
+                image_file_name = os.path.join(directory, "out.%06d.png" % next_index)
+                state_file_name = os.path.join(directory, "out.%06d.json" % next_index)
+                im0.save(image_file_name)
+                with open(state_file_name, "wt") as f:
+                    f.write(json.dumps({
+                        "resolution": resolution,
+                        "prompt": prompt,
+                        "quality": quality,
+                        "style": style
+                    }, indent=2, ensure_ascii=False))
+
             im1 = TF.to_tensor(im0.convert("RGBA"))
             im1[:3, im1[3, :, :] == 0] = 0
             revised_prompt = r0.data[0].revised_prompt

negi/openai_gpt.py

@@ -0,0 +1,89 @@
+import openai
+import time
+import urllib.error
+
+
+class OpenAiGpt:
+    def __init__(self):
+        self.__client = openai.OpenAI()
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "model": ([
+                    "gpt-4o",
+                    "gpt-4o-2024-05-13",
+                    "gpt-4-turbo",
+                    "gpt-4-turbo-2024-04-09",
+                    "gpt-4-turbo-preview",
+                    "gpt-4-0125-preview",
+                    "gpt-4-1106-preview",
+                    "gpt-4-vision-preview",
+                    "gpt-4",
+                    "gpt-4-0314",
+                    "gpt-4-0613",
+                    "gpt-4-32k",
+                    "gpt-4-32k-0314",
+                    "gpt-4-32k-0613",
+                    "gpt-3.5-turbo",
+                    "gpt-3.5-turbo-16k",
+                    "gpt-3.5-turbo-0301",
+                    "gpt-3.5-turbo-0613",
+                    "gpt-3.5-turbo-1106",
+                    "gpt-3.5-turbo-16k-0613",
+                ], {"default": "gpt-4-0613"}),
+                "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
+                "max_tokens": ("INT", {"default": 512, "min": 16, "max": 4096}),
+                "system_prompt": ("STRING", {
+                    "multiline": True,
+                    "default": "You are a helpful assistant."
+                })
+            },
+            "optional": {
+                "a_role": (["user", "assistant", "system"], {"default": "user"}),
+                "a": ("STRING", {"multiline": False, "default": ""}),
+                "b_role": (["user", "assistant", "system"], {"default": "assistant"}),
+                "b": ("STRING", {"multiline": False, "default": ""}),
+                "c_role": (["user", "assistant", "system"], {"default": "user"}),
+                "c": ("STRING", {"multiline": False, "default": ""}),
+            }
+        }
+
+    RETURN_TYPES = ("STRING",)
+    FUNCTION = "doit"
+    OUTPUT_NODE = False
+    CATEGORY = "Generator"
+
+    def doit(self, model, seed, max_tokens, system_prompt, a_role, a, b_role, b, c_role, c):
+        messages = [{"role": "system", "content": system_prompt}]
+        input_role = [a_role, b_role, c_role]
+        input_text = [a, b, c]
+        for i in range(3):
+            if input_text[i] is not None and len(input_text[i]) > 0:
+                messages.append({
+                    "role": input_role[i] if input_role[i] is not None else "user",
+                    "content": input_text[i]
+                })
+
+        try_count = 0
+        r0 = None
+        while True:
+            try_count += 1
+            try:
+                r0 = self.__client.chat.completions.create(
+                    model=model,
+                    max_tokens=max_tokens,
+                    seed=seed,
+                    messages=messages
+                )
+                break
+            except openai.AuthenticationError as ex:
+                raise ex
+            except (urllib.error.HTTPError, openai.OpenAIError) as ex:
+                if try_count >= 3:
+                    raise ex
+                time.sleep(5)
+                continue
+
+        return (r0.choices[0].message.content,)

negi/openai_gpt4v.py

@@ -0,0 +1,91 @@
+import os
+import base64
+
+import openai
+import requests
+
+import torch
+import torchvision
+
+
+_api_key = os.environ.get("OPENAI_API_KEY")
+_tmp_file = "gpt4v_tmp.jpg"
+
+
+class OpenAiGpt4v:
+    def __init__(self):
+        pass
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
+                "model": ([
+                    "gpt-4o",
+                    "gpt-4o-mini",
+                    "gpt-4-turbo",
+                    "gpt-4-vision-preview",
+                ], {"default": "gpt-4o"}),
+                "detail": (["auto", "low", "high"],),
+                "max_tokens": ("INT", {"default": 512, "min": 16, "max": 4096}),
+                "prompt": ("STRING", {
+                    "multiline": True,
+                    "default": "What’s in this image?"
+                }),
+            }
+        }
+
+    RETURN_TYPES = ("STRING",)
+    FUNCTION = "doit"
+    OUTPUT_NODE = False
+    CATEGORY = "Generator"
+
+    def doit(self, image, seed, model, detail, max_tokens, prompt):
+        _ = seed
+
+        im0 = torchvision.transforms.functional.to_pil_image(torch.permute(image[0], (2, 0, 1)))
+        im0.save(_tmp_file)
+        with open(_tmp_file, "rb") as f:
+            encoded_image = base64.b64encode(f.read()).decode("utf-8")
+
+        headers = {
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {_api_key}"
+        }
+
+        payload = {
+            "model": model if model is not None and model != "" else "gpt-4o",
+            "messages": [
+                {
+                    "role": "user",
+                    "content": [
+                        {
+                            "type": "text",
+                            "text": prompt
+                        },
+                        {
+                            "type": "image_url",
+                            "image_url": {
+                                "url": f"data:image/jpeg;base64,{encoded_image}",
+                                "detail": detail
+                            }
+                        }
+                    ]
+                }
+            ],
+            "max_tokens": max_tokens
+        }
+
+        r0 = requests.post("https://api.openai.com/v1/chat/completions", headers=headers, json=payload)
+        if r0.status_code != 200:
+            raise openai.BadRequestError("Server returned an error", body=None, response=r0)
+
+        r1 = r0.json()
+        if "choices" not in r1 or len(r1["choices"]) < 1:
+            raise openai.BadRequestError("Empty results returned", body=None, response=r0)
+        r2 = r1["choices"][0]
+        if "finish_reason" not in r2 or r2["finish_reason"] != "stop":
+            raise openai.BadRequestError("Request was not completed correctly", body=None, response=r0)
+        return (r2["message"]["content"],)

negi/repos/controlnet_aux/.gitignore

@@ -0,0 +1,178 @@
+# Initially taken from Github's Python gitignore file
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# tests and logs
+tests/fixtures/cached_*_text.txt
+logs/
+lightning_logs/
+lang_code_data/
+tests/outputs
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+.python-version
+
+# celery beat schedule file
+celerybeat-schedule
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# vscode
+.vs
+.vscode
+
+# Pycharm
+.idea
+
+# TF code
+tensorflow_code
+
+# Models
+proc_data
+
+# examples
+runs
+/runs_old
+/wandb
+/examples/runs
+/examples/**/*.args
+/examples/rag/sweep
+
+# data
+/data
+serialization_dir
+
+# emacs
+*.*~
+debug.env
+
+# vim
+.*.swp
+
+#ctags
+tags
+
+# pre-commit
+.pre-commit*
+
+# .lock
+*.lock
+
+# DS_Store (MacOS)
+.DS_Store
+# RL pipelines may produce mp4 outputs
+*.mp4
+
+# dependencies
+/transformers
+
+# ruff
+.ruff_cache
+
+wandb
+

negi/repos/controlnet_aux/LICENSE.txt

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

negi/repos/controlnet_aux/README.md

@@ -0,0 +1,112 @@
+# ControlNet auxiliary models
+
+This is a PyPi installable package of [lllyasviel's ControlNet Annotators](https://github.com/lllyasviel/ControlNet/tree/main/annotator)
+
+The code is copy-pasted from the respective folders in https://github.com/lllyasviel/ControlNet/tree/main/annotator and connected to [the 🤗 Hub](https://huggingface.co/lllyasviel/Annotators).
+
+All credit & copyright goes to https://github.com/lllyasviel .
+
+## Install
+
+```
+pip install controlnet-aux==0.0.6
+```
+
+To support DWPose which is dependent on MMDetection, MMCV and MMPose
+```
+pip install -U openmim
+mim install mmengine
+mim install "mmcv>=2.0.1"
+mim install "mmdet>=3.1.0"
+mim install "mmpose>=1.1.0"
+```
+## Usage
+
+
+You can use the processor class, which can load each of the auxiliary models with the following code
+```python
+import requests
+from PIL import Image
+from io import BytesIO
+
+from controlnet_aux.processor import Processor
+
+# load image
+url = "https://huggingface.co/lllyasviel/sd-controlnet-openpose/resolve/main/images/pose.png"
+
+response = requests.get(url)
+img = Image.open(BytesIO(response.content)).convert("RGB").resize((512, 512))
+
+# load processor from processor_id
+# options are:
+# ["canny", "depth_leres", "depth_leres++", "depth_midas", "depth_zoe", "lineart_anime",
+#  "lineart_coarse", "lineart_realistic", "mediapipe_face", "mlsd", "normal_bae", "normal_midas",
+#  "openpose", "openpose_face", "openpose_faceonly", "openpose_full", "openpose_hand",
+#  "scribble_hed, "scribble_pidinet", "shuffle", "softedge_hed", "softedge_hedsafe",
+#  "softedge_pidinet", "softedge_pidsafe", "dwpose"]
+processor_id = 'scribble_hed'
+processor = Processor(processor_id)
+
+processed_image = processor(img, to_pil=True)
+```
+
+Each model can be loaded individually by importing and instantiating them as follows
+```python
+from PIL import Image
+import requests
+from io import BytesIO
+from controlnet_aux import HEDdetector, MidasDetector, MLSDdetector, OpenposeDetector, PidiNetDetector, NormalBaeDetector, LineartDetector, LineartAnimeDetector, CannyDetector, ContentShuffleDetector, ZoeDetector, MediapipeFaceDetector, SamDetector, LeresDetector, DWposeDetector
+
+# load image
+url = "https://huggingface.co/lllyasviel/sd-controlnet-openpose/resolve/main/images/pose.png"
+
+response = requests.get(url)
+img = Image.open(BytesIO(response.content)).convert("RGB").resize((512, 512))
+
+# load checkpoints
+hed = HEDdetector.from_pretrained("lllyasviel/Annotators")
+midas = MidasDetector.from_pretrained("lllyasviel/Annotators")
+mlsd = MLSDdetector.from_pretrained("lllyasviel/Annotators")
+open_pose = OpenposeDetector.from_pretrained("lllyasviel/Annotators")
+pidi = PidiNetDetector.from_pretrained("lllyasviel/Annotators")
+normal_bae = NormalBaeDetector.from_pretrained("lllyasviel/Annotators")
+lineart = LineartDetector.from_pretrained("lllyasviel/Annotators")
+lineart_anime = LineartAnimeDetector.from_pretrained("lllyasviel/Annotators")
+zoe = ZoeDetector.from_pretrained("lllyasviel/Annotators")
+sam = SamDetector.from_pretrained("ybelkada/segment-anything", subfolder="checkpoints")
+mobile_sam = SamDetector.from_pretrained("dhkim2810/MobileSAM", model_type="vit_t", filename="mobile_sam.pt")
+leres = LeresDetector.from_pretrained("lllyasviel/Annotators")
+
+# specify configs, ckpts and device, or it will be downloaded automatically and use cpu by default
+# det_config: ./src/controlnet_aux/dwpose/yolox_config/yolox_l_8xb8-300e_coco.py
+# det_ckpt: https://download.openmmlab.com/mmdetection/v2.0/yolox/yolox_l_8x8_300e_coco/yolox_l_8x8_300e_coco_20211126_140236-d3bd2b23.pth
+# pose_config: ./src/controlnet_aux/dwpose/dwpose_config/dwpose-l_384x288.py
+# pose_ckpt: https://huggingface.co/wanghaofan/dw-ll_ucoco_384/resolve/main/dw-ll_ucoco_384.pth
+import torch
+device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+dwpose = DWposeDetector(det_config=det_config, det_ckpt=det_ckpt, pose_config=pose_config, pose_ckpt=pose_ckpt, device=device)
+
+# instantiate
+canny = CannyDetector()
+content = ContentShuffleDetector()
+face_detector = MediapipeFaceDetector()
+
+
+# process
+processed_image_hed = hed(img)
+processed_image_midas = midas(img)
+processed_image_mlsd = mlsd(img)
+processed_image_open_pose = open_pose(img, hand_and_face=True)
+processed_image_pidi = pidi(img, safe=True)
+processed_image_normal_bae = normal_bae(img)
+processed_image_lineart = lineart(img, coarse=True)
+processed_image_lineart_anime = lineart_anime(img)
+processed_image_zoe = zoe(img)
+processed_image_sam = sam(img)
+processed_image_leres = leres(img)
+
+processed_image_canny = canny(img)
+processed_image_content = content(img)
+processed_image_mediapipe_face = face_detector(img)
+processed_image_dwpose = dwpose(img)
+```

negi/repos/controlnet_aux/setup.py

@@ -0,0 +1,233 @@
+# Copyright 2023 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+Simple check list from AllenNLP repo: https://github.com/allenai/allennlp/blob/main/setup.py
+
+To create the package for pypi.
+
+1. Run `make pre-release` (or `make pre-patch` for a patch release) then run `make fix-copies` to fix the index of the
+   documentation.
+
+   If releasing on a special branch, copy the updated README.md on the main branch for your the commit you will make
+   for the post-release and run `make fix-copies` on the main branch as well.
+
+2. Run Tests for Amazon Sagemaker. The documentation is located in `./tests/sagemaker/README.md`, otherwise @philschmid.
+
+3. Unpin specific versions from setup.py that use a git install.
+
+4. Checkout the release branch (v<RELEASE>-release, for example v4.19-release), and commit these changes with the
+   message: "Release: <RELEASE>" and push.
+
+5. Wait for the tests on main to be completed and be green (otherwise revert and fix bugs)
+
+6. Add a tag in git to mark the release: "git tag v<RELEASE> -m 'Adds tag v<RELEASE> for pypi' "
+   Push the tag to git: git push --tags origin v<RELEASE>-release
+
+7. Build both the sources and the wheel. Do not change anything in setup.py between
+   creating the wheel and the source distribution (obviously).
+
+   For the wheel, run: "python setup.py bdist_wheel" in the top level directory.
+   (this will build a wheel for the python version you use to build it).
+
+   For the sources, run: "python setup.py sdist"
+   You should now have a /dist directory with both .whl and .tar.gz source versions.
+
+8. Check that everything looks correct by uploading the package to the pypi test server:
+
+   twine upload dist/* -r pypitest
+   (pypi suggest using twine as other methods upload files via plaintext.)
+   You may have to specify the repository url, use the following command then:
+   twine upload dist/* -r pypitest --repository-url=https://test.pypi.org/legacy/
+
+   Check that you can install it in a virtualenv by running:
+   pip install -i https://testpypi.python.org/pypi diffusers
+
+   Check you can run the following commands:
+   python -c "from diffusers import pipeline; classifier = pipeline('text-classification'); print(classifier('What a nice release'))"
+   python -c "from diffusers import *"
+
+9. Upload the final version to actual pypi:
+   twine upload dist/* -r pypi
+
+10. Copy the release notes from RELEASE.md to the tag in github once everything is looking hunky-dory.
+
+11. Run `make post-release` (or, for a patch release, `make post-patch`). If you were on a branch for the release,
+    you need to go back to main before executing this.
+"""
+
+import os
+import re
+from distutils.core import Command
+
+from setuptools import find_packages, setup
+
+# IMPORTANT:
+# 1. all dependencies should be listed here with their version requirements if any
+# 2. once modified, run: `make deps_table_update` to update src/diffusers/dependency_versions_table.py
+_deps = [
+    "Pillow",
+    "torch",
+    "numpy",
+    "filelock",
+    "importlib_metadata",
+    "opencv-python",
+    "scipy",
+    "huggingface_hub",
+    "einops",
+    "timm",
+    "torchvision",
+    "scikit-image"
+]
+
+# this is a lookup table with items like:
+#
+# tokenizers: "huggingface-hub==0.8.0"
+# packaging: "packaging"
+#
+# some of the values are versioned whereas others aren't.
+deps = {
+    b: a for a, b in (re.findall(r"^(([^!=<>~]+)(?:[!=<>~].*)?$)", x)[0] for x in _deps)
+}
+
+# since we save this data in src/diffusers/dependency_versions_table.py it can be easily accessed from
+# anywhere. If you need to quickly access the data from this table in a shell, you can do so easily with:
+#
+# python -c 'import sys; from diffusers.dependency_versions_table import deps; \
+# print(" ".join([ deps[x] for x in sys.argv[1:]]))' tokenizers datasets
+#
+# Just pass the desired package names to that script as it's shown with 2 packages above.
+#
+# If diffusers is not yet installed and the work is done from the cloned repo remember to add `PYTHONPATH=src` to the script above
+#
+# You can then feed this for example to `pip`:
+#
+# pip install -U $(python -c 'import sys; from diffusers.dependency_versions_table import deps; \
+# print(" ".join([ deps[x] for x in sys.argv[1:]]))' tokenizers datasets)
+#
+
+
+def deps_list(*pkgs):
+    return [deps[pkg] for pkg in pkgs]
+
+
+class DepsTableUpdateCommand(Command):
+    """
+    A custom distutils command that updates the dependency table.
+    usage: python setup.py deps_table_update
+    """
+
+    description = "build runtime dependency table"
+    user_options = [
+        # format: (long option, short option, description).
+        (
+            "dep-table-update",
+            None,
+            "updates src/diffusers/dependency_versions_table.py",
+        ),
+    ]
+
+    def initialize_options(self):
+        pass
+
+    def finalize_options(self):
+        pass
+
+    def run(self):
+        entries = "\n".join([f'    "{k}": "{v}",' for k, v in deps.items()])
+        content = [
+            "# THIS FILE HAS BEEN AUTOGENERATED. To update:",
+            "# 1. modify the `_deps` dict in setup.py",
+            "# 2. run `make deps_table_update``",
+            "deps = {",
+            entries,
+            "}",
+            "",
+        ]
+        target = "src/controlnet_aux/dependency_versions_table.py"
+        print(f"updating {target}")
+        with open(target, "w", encoding="utf-8", newline="\n") as f:
+            f.write("\n".join(content))
+
+
+extras = {}
+
+install_requires = [
+    deps["torch"],
+    deps["importlib_metadata"],
+    deps["huggingface_hub"],
+    deps["scipy"],
+    deps["opencv-python"],
+    deps["filelock"],
+    deps["numpy"],
+    deps["Pillow"],
+    deps["einops"],
+    deps["torchvision"],
+    deps["timm"],
+    deps["scikit-image"],
+]
+
+setup(
+    name="controlnet_aux",
+    version="0.0.7",  # expected format is one of x.y.z.dev0, or x.y.z.rc1 or x.y.z (no to dashes, yes to dots)
+    description="Auxillary models for controlnet",
+    long_description=open("README.md", "r", encoding="utf-8").read(),
+    long_description_content_type="text/markdown",
+    keywords="deep learning",
+    license="Apache",
+    author="The HuggingFace team",
+    author_email="patrick@huggingface.co",
+    url="https://github.com/patrickvonplaten/controlnet_aux",
+    package_dir={"": "src"},
+    packages=find_packages("src"),
+    include_package_data=True,
+    python_requires=">=3.7.0",
+    install_requires=install_requires,
+    extras_require=extras,
+    classifiers=[
+        "Development Status :: 5 - Production/Stable",
+        "Intended Audience :: Developers",
+        "Intended Audience :: Education",
+        "Intended Audience :: Science/Research",
+        "License :: OSI Approved :: Apache Software License",
+        "Operating System :: OS Independent",
+        "Programming Language :: Python :: 3",
+        "Programming Language :: Python :: 3.7",
+        "Programming Language :: Python :: 3.8",
+        "Programming Language :: Python :: 3.9",
+        "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    ],
+    cmdclass={"deps_table_update": DepsTableUpdateCommand},
+    package_data={'controlnet_aux' : ['zoe/zoedepth/models/zoedepth/*.json', 'zoe/zoedepth/models/zoedepth_nk/*.json']}
+)
+
+# Release checklist
+# 1. Change the version in __init__.py and setup.py.
+# 2. Commit these changes with the message: "Release: Release"
+# 3. Add a tag in git to mark the release: "git tag RELEASE -m 'Adds tag RELEASE for pypi' "
+#    Push the tag to git: git push --tags origin main
+# 4. Run the following commands in the top-level directory:
+#      python setup.py bdist_wheel
+#      python setup.py sdist
+# 5. Upload the package to the pypi test server first:
+#      twine upload dist/* -r pypitest
+#      twine upload dist/* -r pypitest --repository-url=https://test.pypi.org/legacy/
+# 6. Check that you can install it in a virtualenv by running:
+#      pip install -i https://testpypi.python.org/pypi diffusers
+#      diffusers env
+#      diffusers test
+# 7. Upload the final version to actual pypi:
+#      twine upload dist/* -r pypi
+# 8. Add release notes to the tag in github once everything is looking hunky-dory.
+# 9. Update the version in __init__.py, setup.py to the new version "-dev" and push to master

negi/repos/controlnet_aux/src/controlnet_aux/__init__.py

@@ -0,0 +1,18 @@
+__version__ = "0.0.7"
+
+# from .hed import HEDdetector
+# from .leres import LeresDetector
+# from .lineart import LineartDetector
+# from .lineart_anime import LineartAnimeDetector
+# from .midas import MidasDetector
+# from .mlsd import MLSDdetector
+# from .normalbae import NormalBaeDetector
+from .open_pose import OpenposeDetector
+# from .pidi import PidiNetDetector
+# from .zoe import ZoeDetector
+
+# from .canny import CannyDetector
+# from .mediapipe_face import MediapipeFaceDetector
+# from .segment_anything import SamDetector
+# from .shuffle import ContentShuffleDetector
+# from .dwpose import DWposeDetector

negi/repos/controlnet_aux/src/controlnet_aux/canny/__init__.py

@@ -0,0 +1,36 @@
+import warnings
+import cv2
+import numpy as np
+from PIL import Image
+from ..util import HWC3, resize_image
+
+class CannyDetector:
+    def __call__(self, input_image=None, low_threshold=100, high_threshold=200, detect_resolution=512, image_resolution=512, output_type=None, **kwargs):
+        if "img" in kwargs:
+            warnings.warn("img is deprecated, please use `input_image=...` instead.", DeprecationWarning)
+            input_image = kwargs.pop("img")
+        
+        if input_image is None:
+            raise ValueError("input_image must be defined.")
+
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+            output_type = output_type or "pil"
+        else:
+            output_type = output_type or "np"
+        
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+
+        detected_map = cv2.Canny(input_image, low_threshold, high_threshold)
+        detected_map = HWC3(detected_map)      
+         
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+        
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+            
+        return detected_map

negi/repos/controlnet_aux/src/controlnet_aux/dwpose/__init__.py

@@ -0,0 +1,91 @@
+# Openpose
+# Original from CMU https://github.com/CMU-Perceptual-Computing-Lab/openpose
+# 2nd Edited by https://github.com/Hzzone/pytorch-openpose
+# 3rd Edited by ControlNet
+# 4th Edited by ControlNet (added face and correct hands)
+
+import os
+os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
+
+import cv2
+import torch
+import numpy as np
+from PIL import Image
+
+from ..util import HWC3, resize_image
+from . import util
+
+
+def draw_pose(pose, H, W):
+    bodies = pose['bodies']
+    faces = pose['faces']
+    hands = pose['hands']
+    candidate = bodies['candidate']
+    subset = bodies['subset']
+    
+    canvas = np.zeros(shape=(H, W, 3), dtype=np.uint8)
+    canvas = util.draw_bodypose(canvas, candidate, subset)
+    canvas = util.draw_handpose(canvas, hands)
+    canvas = util.draw_facepose(canvas, faces)
+
+    return canvas
+
+class DWposeDetector:
+    def __init__(self, det_config=None, det_ckpt=None, pose_config=None, pose_ckpt=None, device="cpu"):
+        from .wholebody import Wholebody
+
+        self.pose_estimation = Wholebody(det_config, det_ckpt, pose_config, pose_ckpt, device)
+    
+    def to(self, device):
+        self.pose_estimation.to(device)
+        return self
+    
+    def __call__(self, input_image, detect_resolution=512, image_resolution=512, output_type="pil", **kwargs):
+        
+        input_image = cv2.cvtColor(np.array(input_image, dtype=np.uint8), cv2.COLOR_RGB2BGR)
+
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+        H, W, C = input_image.shape
+        
+        with torch.no_grad():
+            candidate, subset = self.pose_estimation(input_image)
+            nums, keys, locs = candidate.shape
+            candidate[..., 0] /= float(W)
+            candidate[..., 1] /= float(H)
+            body = candidate[:,:18].copy()
+            body = body.reshape(nums*18, locs)
+            score = subset[:,:18]
+            
+            for i in range(len(score)):
+                for j in range(len(score[i])):
+                    if score[i][j] > 0.3:
+                        score[i][j] = int(18*i+j)
+                    else:
+                        score[i][j] = -1
+
+            un_visible = subset<0.3
+            candidate[un_visible] = -1
+
+            foot = candidate[:,18:24]
+
+            faces = candidate[:,24:92]
+
+            hands = candidate[:,92:113]
+            hands = np.vstack([hands, candidate[:,113:]])
+            
+            bodies = dict(candidate=body, subset=score)
+            pose = dict(bodies=bodies, hands=hands, faces=faces)
+            
+            detected_map = draw_pose(pose, H, W)
+            detected_map = HWC3(detected_map)
+            
+            img = resize_image(input_image, image_resolution)
+            H, W, C = img.shape
+
+            detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+
+            if output_type == "pil":
+                detected_map = Image.fromarray(detected_map)
+                
+            return detected_map

negi/repos/controlnet_aux/src/controlnet_aux/dwpose/dwpose_config/dwpose-l_384x288.py

@@ -0,0 +1,257 @@
+# runtime
+max_epochs = 270
+stage2_num_epochs = 30
+base_lr = 4e-3
+
+train_cfg = dict(max_epochs=max_epochs, val_interval=10)
+randomness = dict(seed=21)
+
+# optimizer
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=dict(type='AdamW', lr=base_lr, weight_decay=0.05),
+    paramwise_cfg=dict(
+        norm_decay_mult=0, bias_decay_mult=0, bypass_duplicate=True))
+
+# learning rate
+param_scheduler = [
+    dict(
+        type='LinearLR',
+        start_factor=1.0e-5,
+        by_epoch=False,
+        begin=0,
+        end=1000),
+    dict(
+        # use cosine lr from 150 to 300 epoch
+        type='CosineAnnealingLR',
+        eta_min=base_lr * 0.05,
+        begin=max_epochs // 2,
+        end=max_epochs,
+        T_max=max_epochs // 2,
+        by_epoch=True,
+        convert_to_iter_based=True),
+]
+
+# automatically scaling LR based on the actual training batch size
+auto_scale_lr = dict(base_batch_size=512)
+
+# codec settings
+codec = dict(
+    type='SimCCLabel',
+    input_size=(288, 384),
+    sigma=(6., 6.93),
+    simcc_split_ratio=2.0,
+    normalize=False,
+    use_dark=False)
+
+# model settings
+model = dict(
+    type='TopdownPoseEstimator',
+    data_preprocessor=dict(
+        type='PoseDataPreprocessor',
+        mean=[123.675, 116.28, 103.53],
+        std=[58.395, 57.12, 57.375],
+        bgr_to_rgb=True),
+    backbone=dict(
+        _scope_='mmdet',
+        type='CSPNeXt',
+        arch='P5',
+        expand_ratio=0.5,
+        deepen_factor=1.,
+        widen_factor=1.,
+        out_indices=(4, ),
+        channel_attention=True,
+        norm_cfg=dict(type='SyncBN'),
+        act_cfg=dict(type='SiLU'),
+        init_cfg=dict(
+            type='Pretrained',
+            prefix='backbone.',
+            checkpoint='https://download.openmmlab.com/mmpose/v1/projects/'
+            'rtmpose/cspnext-l_udp-aic-coco_210e-256x192-273b7631_20230130.pth'  # noqa
+        )),
+    head=dict(
+        type='RTMCCHead',
+        in_channels=1024,
+        out_channels=133,
+        input_size=codec['input_size'],
+        in_featuremap_size=(9, 12),
+        simcc_split_ratio=codec['simcc_split_ratio'],
+        final_layer_kernel_size=7,
+        gau_cfg=dict(
+            hidden_dims=256,
+            s=128,
+            expansion_factor=2,
+            dropout_rate=0.,
+            drop_path=0.,
+            act_fn='SiLU',
+            use_rel_bias=False,
+            pos_enc=False),
+        loss=dict(
+            type='KLDiscretLoss',
+            use_target_weight=True,
+            beta=10.,
+            label_softmax=True),
+        decoder=codec),
+    test_cfg=dict(flip_test=True, ))
+
+# base dataset settings
+dataset_type = 'CocoWholeBodyDataset'
+data_mode = 'topdown'
+data_root = '/data/'
+
+backend_args = dict(backend='local')
+# backend_args = dict(
+#     backend='petrel',
+#     path_mapping=dict({
+#         f'{data_root}': 's3://openmmlab/datasets/detection/coco/',
+#         f'{data_root}': 's3://openmmlab/datasets/detection/coco/'
+#     }))
+
+# pipelines
+train_pipeline = [
+    dict(type='LoadImage', backend_args=backend_args),
+    dict(type='GetBBoxCenterScale'),
+    dict(type='RandomFlip', direction='horizontal'),
+    dict(type='RandomHalfBody'),
+    dict(
+        type='RandomBBoxTransform', scale_factor=[0.6, 1.4], rotate_factor=80),
+    dict(type='TopdownAffine', input_size=codec['input_size']),
+    dict(type='mmdet.YOLOXHSVRandomAug'),
+    dict(
+        type='Albumentation',
+        transforms=[
+            dict(type='Blur', p=0.1),
+            dict(type='MedianBlur', p=0.1),
+            dict(
+                type='CoarseDropout',
+                max_holes=1,
+                max_height=0.4,
+                max_width=0.4,
+                min_holes=1,
+                min_height=0.2,
+                min_width=0.2,
+                p=1.0),
+        ]),
+    dict(type='GenerateTarget', encoder=codec),
+    dict(type='PackPoseInputs')
+]
+val_pipeline = [
+    dict(type='LoadImage', backend_args=backend_args),
+    dict(type='GetBBoxCenterScale'),
+    dict(type='TopdownAffine', input_size=codec['input_size']),
+    dict(type='PackPoseInputs')
+]
+
+train_pipeline_stage2 = [
+    dict(type='LoadImage', backend_args=backend_args),
+    dict(type='GetBBoxCenterScale'),
+    dict(type='RandomFlip', direction='horizontal'),
+    dict(type='RandomHalfBody'),
+    dict(
+        type='RandomBBoxTransform',
+        shift_factor=0.,
+        scale_factor=[0.75, 1.25],
+        rotate_factor=60),
+    dict(type='TopdownAffine', input_size=codec['input_size']),
+    dict(type='mmdet.YOLOXHSVRandomAug'),
+    dict(
+        type='Albumentation',
+        transforms=[
+            dict(type='Blur', p=0.1),
+            dict(type='MedianBlur', p=0.1),
+            dict(
+                type='CoarseDropout',
+                max_holes=1,
+                max_height=0.4,
+                max_width=0.4,
+                min_holes=1,
+                min_height=0.2,
+                min_width=0.2,
+                p=0.5),
+        ]),
+    dict(type='GenerateTarget', encoder=codec),
+    dict(type='PackPoseInputs')
+]
+
+datasets = []
+dataset_coco=dict(
+    type=dataset_type,
+    data_root=data_root,
+    data_mode=data_mode,
+    ann_file='coco/annotations/coco_wholebody_train_v1.0.json',
+    data_prefix=dict(img='coco/train2017/'),
+    pipeline=[],
+)
+datasets.append(dataset_coco)
+
+scene = ['Magic_show', 'Entertainment', 'ConductMusic', 'Online_class', 
+         'TalkShow', 'Speech', 'Fitness', 'Interview', 'Olympic', 'TVShow', 
+         'Singing', 'SignLanguage', 'Movie', 'LiveVlog', 'VideoConference']
+
+for i in range(len(scene)):
+    datasets.append(
+        dict(
+            type=dataset_type,
+            data_root=data_root,
+            data_mode=data_mode,
+            ann_file='UBody/annotations/'+scene[i]+'/keypoint_annotation.json',
+            data_prefix=dict(img='UBody/images/'+scene[i]+'/'),
+            pipeline=[],
+        )
+    )
+
+# data loaders
+train_dataloader = dict(
+    batch_size=32,
+    num_workers=10,
+    persistent_workers=True,
+    sampler=dict(type='DefaultSampler', shuffle=True),
+    dataset=dict(
+        type='CombinedDataset',
+        metainfo=dict(from_file='configs/_base_/datasets/coco_wholebody.py'),
+        datasets=datasets,
+        pipeline=train_pipeline,
+        test_mode=False,
+    ))
+val_dataloader = dict(
+    batch_size=32,
+    num_workers=10,
+    persistent_workers=True,
+    drop_last=False,
+    sampler=dict(type='DefaultSampler', shuffle=False, round_up=False),
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        data_mode=data_mode,
+        ann_file='coco/annotations/coco_wholebody_val_v1.0.json',
+        bbox_file=f'{data_root}coco/person_detection_results/'
+        'COCO_val2017_detections_AP_H_56_person.json',
+        data_prefix=dict(img='coco/val2017/'),
+        test_mode=True,
+        pipeline=val_pipeline,
+    ))
+test_dataloader = val_dataloader
+
+# hooks
+default_hooks = dict(
+    checkpoint=dict(
+        save_best='coco-wholebody/AP', rule='greater', max_keep_ckpts=1))
+
+custom_hooks = [
+    dict(
+        type='EMAHook',
+        ema_type='ExpMomentumEMA',
+        momentum=0.0002,
+        update_buffers=True,
+        priority=49),
+    dict(
+        type='mmdet.PipelineSwitchHook',
+        switch_epoch=max_epochs - stage2_num_epochs,
+        switch_pipeline=train_pipeline_stage2)
+]
+
+# evaluators
+val_evaluator = dict(
+    type='CocoWholeBodyMetric',
+    ann_file=data_root + 'coco/annotations/coco_wholebody_val_v1.0.json')
+test_evaluator = val_evaluator

negi/repos/controlnet_aux/src/controlnet_aux/dwpose/util.py

@@ -0,0 +1,303 @@
+import math
+import numpy as np
+import cv2
+
+
+eps = 0.01
+
+
+def smart_resize(x, s):
+    Ht, Wt = s
+    if x.ndim == 2:
+        Ho, Wo = x.shape
+        Co = 1
+    else:
+        Ho, Wo, Co = x.shape
+    if Co == 3 or Co == 1:
+        k = float(Ht + Wt) / float(Ho + Wo)
+        return cv2.resize(x, (int(Wt), int(Ht)), interpolation=cv2.INTER_AREA if k < 1 else cv2.INTER_LANCZOS4)
+    else:
+        return np.stack([smart_resize(x[:, :, i], s) for i in range(Co)], axis=2)
+
+
+def smart_resize_k(x, fx, fy):
+    if x.ndim == 2:
+        Ho, Wo = x.shape
+        Co = 1
+    else:
+        Ho, Wo, Co = x.shape
+    Ht, Wt = Ho * fy, Wo * fx
+    if Co == 3 or Co == 1:
+        k = float(Ht + Wt) / float(Ho + Wo)
+        return cv2.resize(x, (int(Wt), int(Ht)), interpolation=cv2.INTER_AREA if k < 1 else cv2.INTER_LANCZOS4)
+    else:
+        return np.stack([smart_resize_k(x[:, :, i], fx, fy) for i in range(Co)], axis=2)
+
+
+def padRightDownCorner(img, stride, padValue):
+    h = img.shape[0]
+    w = img.shape[1]
+
+    pad = 4 * [None]
+    pad[0] = 0 # up
+    pad[1] = 0 # left
+    pad[2] = 0 if (h % stride == 0) else stride - (h % stride) # down
+    pad[3] = 0 if (w % stride == 0) else stride - (w % stride) # right
+
+    img_padded = img
+    pad_up = np.tile(img_padded[0:1, :, :]*0 + padValue, (pad[0], 1, 1))
+    img_padded = np.concatenate((pad_up, img_padded), axis=0)
+    pad_left = np.tile(img_padded[:, 0:1, :]*0 + padValue, (1, pad[1], 1))
+    img_padded = np.concatenate((pad_left, img_padded), axis=1)
+    pad_down = np.tile(img_padded[-2:-1, :, :]*0 + padValue, (pad[2], 1, 1))
+    img_padded = np.concatenate((img_padded, pad_down), axis=0)
+    pad_right = np.tile(img_padded[:, -2:-1, :]*0 + padValue, (1, pad[3], 1))
+    img_padded = np.concatenate((img_padded, pad_right), axis=1)
+
+    return img_padded, pad
+
+
+def transfer(model, model_weights):
+    transfered_model_weights = {}
+    for weights_name in model.state_dict().keys():
+        transfered_model_weights[weights_name] = model_weights['.'.join(weights_name.split('.')[1:])]
+    return transfered_model_weights
+
+
+def draw_bodypose(canvas, candidate, subset):
+    H, W, C = canvas.shape
+    candidate = np.array(candidate)
+    subset = np.array(subset)
+
+    stickwidth = 4
+
+    limbSeq = [[2, 3], [2, 6], [3, 4], [4, 5], [6, 7], [7, 8], [2, 9], [9, 10], \
+               [10, 11], [2, 12], [12, 13], [13, 14], [2, 1], [1, 15], [15, 17], \
+               [1, 16], [16, 18], [3, 17], [6, 18]]
+
+    colors = [[255, 0, 0], [255, 85, 0], [255, 170, 0], [255, 255, 0], [170, 255, 0], [85, 255, 0], [0, 255, 0], \
+              [0, 255, 85], [0, 255, 170], [0, 255, 255], [0, 170, 255], [0, 85, 255], [0, 0, 255], [85, 0, 255], \
+              [170, 0, 255], [255, 0, 255], [255, 0, 170], [255, 0, 85]]
+
+    for i in range(17):
+        for n in range(len(subset)):
+            index = subset[n][np.array(limbSeq[i]) - 1]
+            if -1 in index:
+                continue
+            Y = candidate[index.astype(int), 0] * float(W)
+            X = candidate[index.astype(int), 1] * float(H)
+            mX = np.mean(X)
+            mY = np.mean(Y)
+            length = ((X[0] - X[1]) ** 2 + (Y[0] - Y[1]) ** 2) ** 0.5
+            angle = math.degrees(math.atan2(X[0] - X[1], Y[0] - Y[1]))
+            polygon = cv2.ellipse2Poly((int(mY), int(mX)), (int(length / 2), stickwidth), int(angle), 0, 360, 1)
+            cv2.fillConvexPoly(canvas, polygon, colors[i])
+
+    canvas = (canvas * 0.6).astype(np.uint8)
+
+    for i in range(18):
+        for n in range(len(subset)):
+            index = int(subset[n][i])
+            if index == -1:
+                continue
+            x, y = candidate[index][0:2]
+            x = int(x * W)
+            y = int(y * H)
+            cv2.circle(canvas, (int(x), int(y)), 4, colors[i], thickness=-1)
+
+    return canvas
+
+
+def draw_handpose(canvas, all_hand_peaks):
+    import matplotlib
+    
+    H, W, C = canvas.shape
+
+    edges = [[0, 1], [1, 2], [2, 3], [3, 4], [0, 5], [5, 6], [6, 7], [7, 8], [0, 9], [9, 10], \
+             [10, 11], [11, 12], [0, 13], [13, 14], [14, 15], [15, 16], [0, 17], [17, 18], [18, 19], [19, 20]]
+    
+    # (person_number*2, 21, 2)
+    for i in range(len(all_hand_peaks)):
+        peaks = all_hand_peaks[i]
+        peaks = np.array(peaks)
+        
+        for ie, e in enumerate(edges):
+
+            x1, y1 = peaks[e[0]]
+            x2, y2 = peaks[e[1]]
+            
+            x1 = int(x1 * W)
+            y1 = int(y1 * H)
+            x2 = int(x2 * W)
+            y2 = int(y2 * H)
+            if x1 > eps and y1 > eps and x2 > eps and y2 > eps:
+                cv2.line(canvas, (x1, y1), (x2, y2), matplotlib.colors.hsv_to_rgb([ie / float(len(edges)), 1.0, 1.0]) * 255, thickness=2)
+
+        for _, keyponit in enumerate(peaks):
+            x, y = keyponit
+
+            x = int(x * W)
+            y = int(y * H)
+            if x > eps and y > eps:
+                cv2.circle(canvas, (x, y), 4, (0, 0, 255), thickness=-1)
+    return canvas
+
+
+def draw_facepose(canvas, all_lmks):
+    H, W, C = canvas.shape
+    for lmks in all_lmks:
+        lmks = np.array(lmks)
+        for lmk in lmks:
+            x, y = lmk
+            x = int(x * W)
+            y = int(y * H)
+            if x > eps and y > eps:
+                cv2.circle(canvas, (x, y), 3, (255, 255, 255), thickness=-1)
+    return canvas
+
+
+# detect hand according to body pose keypoints
+# please refer to https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/src/openpose/hand/handDetector.cpp
+def handDetect(candidate, subset, oriImg):
+    # right hand: wrist 4, elbow 3, shoulder 2
+    # left hand: wrist 7, elbow 6, shoulder 5
+    ratioWristElbow = 0.33
+    detect_result = []
+    image_height, image_width = oriImg.shape[0:2]
+    for person in subset.astype(int):
+        # if any of three not detected
+        has_left = np.sum(person[[5, 6, 7]] == -1) == 0
+        has_right = np.sum(person[[2, 3, 4]] == -1) == 0
+        if not (has_left or has_right):
+            continue
+        hands = []
+        #left hand
+        if has_left:
+            left_shoulder_index, left_elbow_index, left_wrist_index = person[[5, 6, 7]]
+            x1, y1 = candidate[left_shoulder_index][:2]
+            x2, y2 = candidate[left_elbow_index][:2]
+            x3, y3 = candidate[left_wrist_index][:2]
+            hands.append([x1, y1, x2, y2, x3, y3, True])
+        # right hand
+        if has_right:
+            right_shoulder_index, right_elbow_index, right_wrist_index = person[[2, 3, 4]]
+            x1, y1 = candidate[right_shoulder_index][:2]
+            x2, y2 = candidate[right_elbow_index][:2]
+            x3, y3 = candidate[right_wrist_index][:2]
+            hands.append([x1, y1, x2, y2, x3, y3, False])
+
+        for x1, y1, x2, y2, x3, y3, is_left in hands:
+            # pos_hand = pos_wrist + ratio * (pos_wrist - pos_elbox) = (1 + ratio) * pos_wrist - ratio * pos_elbox
+            # handRectangle.x = posePtr[wrist*3] + ratioWristElbow * (posePtr[wrist*3] - posePtr[elbow*3]);
+            # handRectangle.y = posePtr[wrist*3+1] + ratioWristElbow * (posePtr[wrist*3+1] - posePtr[elbow*3+1]);
+            # const auto distanceWristElbow = getDistance(poseKeypoints, person, wrist, elbow);
+            # const auto distanceElbowShoulder = getDistance(poseKeypoints, person, elbow, shoulder);
+            # handRectangle.width = 1.5f * fastMax(distanceWristElbow, 0.9f * distanceElbowShoulder);
+            x = x3 + ratioWristElbow * (x3 - x2)
+            y = y3 + ratioWristElbow * (y3 - y2)
+            distanceWristElbow = math.sqrt((x3 - x2) ** 2 + (y3 - y2) ** 2)
+            distanceElbowShoulder = math.sqrt((x2 - x1) ** 2 + (y2 - y1) ** 2)
+            width = 1.5 * max(distanceWristElbow, 0.9 * distanceElbowShoulder)
+            # x-y refers to the center --> offset to topLeft point
+            # handRectangle.x -= handRectangle.width / 2.f;
+            # handRectangle.y -= handRectangle.height / 2.f;
+            x -= width / 2
+            y -= width / 2  # width = height
+            # overflow the image
+            if x < 0: x = 0
+            if y < 0: y = 0
+            width1 = width
+            width2 = width
+            if x + width > image_width: width1 = image_width - x
+            if y + width > image_height: width2 = image_height - y
+            width = min(width1, width2)
+            # the max hand box value is 20 pixels
+            if width >= 20:
+                detect_result.append([int(x), int(y), int(width), is_left])
+
+    '''
+    return value: [[x, y, w, True if left hand else False]].
+    width=height since the network require squared input.
+    x, y is the coordinate of top left 
+    '''
+    return detect_result
+
+
+# Written by Lvmin
+def faceDetect(candidate, subset, oriImg):
+    # left right eye ear 14 15 16 17
+    detect_result = []
+    image_height, image_width = oriImg.shape[0:2]
+    for person in subset.astype(int):
+        has_head = person[0] > -1
+        if not has_head:
+            continue
+
+        has_left_eye = person[14] > -1
+        has_right_eye = person[15] > -1
+        has_left_ear = person[16] > -1
+        has_right_ear = person[17] > -1
+
+        if not (has_left_eye or has_right_eye or has_left_ear or has_right_ear):
+            continue
+
+        head, left_eye, right_eye, left_ear, right_ear = person[[0, 14, 15, 16, 17]]
+
+        width = 0.0
+        x0, y0 = candidate[head][:2]
+
+        if has_left_eye:
+            x1, y1 = candidate[left_eye][:2]
+            d = max(abs(x0 - x1), abs(y0 - y1))
+            width = max(width, d * 3.0)
+
+        if has_right_eye:
+            x1, y1 = candidate[right_eye][:2]
+            d = max(abs(x0 - x1), abs(y0 - y1))
+            width = max(width, d * 3.0)
+
+        if has_left_ear:
+            x1, y1 = candidate[left_ear][:2]
+            d = max(abs(x0 - x1), abs(y0 - y1))
+            width = max(width, d * 1.5)
+
+        if has_right_ear:
+            x1, y1 = candidate[right_ear][:2]
+            d = max(abs(x0 - x1), abs(y0 - y1))
+            width = max(width, d * 1.5)
+
+        x, y = x0, y0
+
+        x -= width
+        y -= width
+
+        if x < 0:
+            x = 0
+
+        if y < 0:
+            y = 0
+
+        width1 = width * 2
+        width2 = width * 2
+
+        if x + width > image_width:
+            width1 = image_width - x
+
+        if y + width > image_height:
+            width2 = image_height - y
+
+        width = min(width1, width2)
+
+        if width >= 20:
+            detect_result.append([int(x), int(y), int(width)])
+
+    return detect_result
+
+
+# get max index of 2d array
+def npmax(array):
+    arrayindex = array.argmax(1)
+    arrayvalue = array.max(1)
+    i = arrayvalue.argmax()
+    j = arrayindex[i]
+    return i, j

negi/repos/controlnet_aux/src/controlnet_aux/dwpose/wholebody.py

@@ -0,0 +1,121 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import os
+import numpy as np
+import warnings
+
+try:
+    import mmcv
+except ImportError:
+    warnings.warn(
+        "The module 'mmcv' is not installed. The package will have limited functionality. Please install it using the command: mim install 'mmcv>=2.0.1'"
+    )
+
+try:
+    from mmpose.apis import inference_topdown
+    from mmpose.apis import init_model as init_pose_estimator
+    from mmpose.evaluation.functional import nms
+    from mmpose.utils import adapt_mmdet_pipeline
+    from mmpose.structures import merge_data_samples
+except ImportError:
+    warnings.warn(
+        "The module 'mmpose' is not installed. The package will have limited functionality. Please install it using the command: mim install 'mmpose>=1.1.0'"
+    )
+        
+try:
+    from mmdet.apis import inference_detector, init_detector
+except ImportError:
+    warnings.warn(
+        "The module 'mmdet' is not installed. The package will have limited functionality. Please install it using the command: mim install 'mmdet>=3.1.0'"
+    )
+
+
+class Wholebody:
+    def __init__(self, 
+                 det_config=None, det_ckpt=None, 
+                 pose_config=None, pose_ckpt=None,
+                 device="cpu"):
+        
+        if det_config is None:
+            det_config = os.path.join(os.path.dirname(__file__), "yolox_config/yolox_l_8xb8-300e_coco.py")
+        
+        if pose_config is None:
+            pose_config = os.path.join(os.path.dirname(__file__), "dwpose_config/dwpose-l_384x288.py")
+
+        if det_ckpt is None:
+            det_ckpt = 'https://download.openmmlab.com/mmdetection/v2.0/yolox/yolox_l_8x8_300e_coco/yolox_l_8x8_300e_coco_20211126_140236-d3bd2b23.pth'
+        
+        if pose_ckpt is None:
+            pose_ckpt = "https://huggingface.co/wanghaofan/dw-ll_ucoco_384/resolve/main/dw-ll_ucoco_384.pth"
+        
+        # build detector
+        self.detector = init_detector(det_config, det_ckpt, device=device)
+        self.detector.cfg = adapt_mmdet_pipeline(self.detector.cfg)
+
+        # build pose estimator
+        self.pose_estimator = init_pose_estimator(
+            pose_config,
+            pose_ckpt,
+            device=device)
+    
+    def to(self, device):
+        self.detector.to(device)
+        self.pose_estimator.to(device)
+        return self
+    
+    def __call__(self, oriImg):
+        # predict bbox
+        det_result = inference_detector(self.detector, oriImg)
+        pred_instance = det_result.pred_instances.cpu().numpy()
+        bboxes = np.concatenate(
+            (pred_instance.bboxes, pred_instance.scores[:, None]), axis=1)
+        bboxes = bboxes[np.logical_and(pred_instance.labels == 0,
+                                    pred_instance.scores > 0.5)]
+    
+        # set NMS threshold
+        bboxes = bboxes[nms(bboxes, 0.7), :4]
+
+        # predict keypoints
+        if len(bboxes) == 0:
+            pose_results = inference_topdown(self.pose_estimator, oriImg)
+        else:
+            pose_results = inference_topdown(self.pose_estimator, oriImg, bboxes)
+        preds = merge_data_samples(pose_results)
+        preds = preds.pred_instances
+
+        # preds = pose_results[0].pred_instances
+        keypoints = preds.get('transformed_keypoints',
+                                        preds.keypoints)
+        if 'keypoint_scores' in preds:
+            scores = preds.keypoint_scores
+        else:
+            scores = np.ones(keypoints.shape[:-1])
+
+        if 'keypoints_visible' in preds:
+            visible = preds.keypoints_visible
+        else:
+            visible = np.ones(keypoints.shape[:-1])
+        keypoints_info = np.concatenate(
+            (keypoints, scores[..., None], visible[..., None]),
+            axis=-1)
+        # compute neck joint
+        neck = np.mean(keypoints_info[:, [5, 6]], axis=1)
+        # neck score when visualizing pred
+        neck[:, 2:4] = np.logical_and(
+            keypoints_info[:, 5, 2:4] > 0.3,
+            keypoints_info[:, 6, 2:4] > 0.3).astype(int)
+        new_keypoints_info = np.insert(
+            keypoints_info, 17, neck, axis=1)
+        mmpose_idx = [
+            17, 6, 8, 10, 7, 9, 12, 14, 16, 13, 15, 2, 1, 4, 3
+        ]
+        openpose_idx = [
+            1, 2, 3, 4, 6, 7, 8, 9, 10, 12, 13, 14, 15, 16, 17
+        ]
+        new_keypoints_info[:, openpose_idx] = \
+            new_keypoints_info[:, mmpose_idx]
+        keypoints_info = new_keypoints_info
+
+        keypoints, scores, visible = keypoints_info[
+            ..., :2], keypoints_info[..., 2], keypoints_info[..., 3]
+        
+        return keypoints, scores

negi/repos/controlnet_aux/src/controlnet_aux/dwpose/yolox_config/yolox_l_8xb8-300e_coco.py

@@ -0,0 +1,245 @@
+img_scale = (640, 640)  # width, height
+
+# model settings
+model = dict(
+    type='YOLOX',
+    data_preprocessor=dict(
+        type='DetDataPreprocessor',
+        pad_size_divisor=32,
+        batch_augments=[
+            dict(
+                type='BatchSyncRandomResize',
+                random_size_range=(480, 800),
+                size_divisor=32,
+                interval=10)
+        ]),
+    backbone=dict(
+        type='CSPDarknet',
+        deepen_factor=1.0,
+        widen_factor=1.0,
+        out_indices=(2, 3, 4),
+        use_depthwise=False,
+        spp_kernal_sizes=(5, 9, 13),
+        norm_cfg=dict(type='BN', momentum=0.03, eps=0.001),
+        act_cfg=dict(type='Swish'),
+    ),
+    neck=dict(
+        type='YOLOXPAFPN',
+        in_channels=[256, 512, 1024],
+        out_channels=256,
+        num_csp_blocks=3,
+        use_depthwise=False,
+        upsample_cfg=dict(scale_factor=2, mode='nearest'),
+        norm_cfg=dict(type='BN', momentum=0.03, eps=0.001),
+        act_cfg=dict(type='Swish')),
+    bbox_head=dict(
+        type='YOLOXHead',
+        num_classes=80,
+        in_channels=256,
+        feat_channels=256,
+        stacked_convs=2,
+        strides=(8, 16, 32),
+        use_depthwise=False,
+        norm_cfg=dict(type='BN', momentum=0.03, eps=0.001),
+        act_cfg=dict(type='Swish'),
+        loss_cls=dict(
+            type='CrossEntropyLoss',
+            use_sigmoid=True,
+            reduction='sum',
+            loss_weight=1.0),
+        loss_bbox=dict(
+            type='IoULoss',
+            mode='square',
+            eps=1e-16,
+            reduction='sum',
+            loss_weight=5.0),
+        loss_obj=dict(
+            type='CrossEntropyLoss',
+            use_sigmoid=True,
+            reduction='sum',
+            loss_weight=1.0),
+        loss_l1=dict(type='L1Loss', reduction='sum', loss_weight=1.0)),
+    train_cfg=dict(assigner=dict(type='SimOTAAssigner', center_radius=2.5)),
+    # In order to align the source code, the threshold of the val phase is
+    # 0.01, and the threshold of the test phase is 0.001.
+    test_cfg=dict(score_thr=0.01, nms=dict(type='nms', iou_threshold=0.65)))
+
+# dataset settings
+data_root = 'data/coco/'
+dataset_type = 'CocoDataset'
+
+# Example to use different file client
+# Method 1: simply set the data root and let the file I/O module
+# automatically infer from prefix (not support LMDB and Memcache yet)
+
+# data_root = 's3://openmmlab/datasets/detection/coco/'
+
+# Method 2: Use `backend_args`, `file_client_args` in versions before 3.0.0rc6
+# backend_args = dict(
+#     backend='petrel',
+#     path_mapping=dict({
+#         './data/': 's3://openmmlab/datasets/detection/',
+#         'data/': 's3://openmmlab/datasets/detection/'
+#     }))
+backend_args = None
+
+train_pipeline = [
+    dict(type='Mosaic', img_scale=img_scale, pad_val=114.0),
+    dict(
+        type='RandomAffine',
+        scaling_ratio_range=(0.1, 2),
+        # img_scale is (width, height)
+        border=(-img_scale[0] // 2, -img_scale[1] // 2)),
+    dict(
+        type='MixUp',
+        img_scale=img_scale,
+        ratio_range=(0.8, 1.6),
+        pad_val=114.0),
+    dict(type='YOLOXHSVRandomAug'),
+    dict(type='RandomFlip', prob=0.5),
+    # According to the official implementation, multi-scale
+    # training is not considered here but in the
+    # 'mmdet/models/detectors/yolox.py'.
+    # Resize and Pad are for the last 15 epochs when Mosaic,
+    # RandomAffine, and MixUp are closed by YOLOXModeSwitchHook.
+    dict(type='Resize', scale=img_scale, keep_ratio=True),
+    dict(
+        type='Pad',
+        pad_to_square=True,
+        # If the image is three-channel, the pad value needs
+        # to be set separately for each channel.
+        pad_val=dict(img=(114.0, 114.0, 114.0))),
+    dict(type='FilterAnnotations', min_gt_bbox_wh=(1, 1), keep_empty=False),
+    dict(type='PackDetInputs')
+]
+
+train_dataset = dict(
+    # use MultiImageMixDataset wrapper to support mosaic and mixup
+    type='MultiImageMixDataset',
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file='annotations/instances_train2017.json',
+        data_prefix=dict(img='train2017/'),
+        pipeline=[
+            dict(type='LoadImageFromFile', backend_args=backend_args),
+            dict(type='LoadAnnotations', with_bbox=True)
+        ],
+        filter_cfg=dict(filter_empty_gt=False, min_size=32),
+        backend_args=backend_args),
+    pipeline=train_pipeline)
+
+test_pipeline = [
+    dict(type='LoadImageFromFile', backend_args=backend_args),
+    dict(type='Resize', scale=img_scale, keep_ratio=True),
+    dict(
+        type='Pad',
+        pad_to_square=True,
+        pad_val=dict(img=(114.0, 114.0, 114.0))),
+    dict(type='LoadAnnotations', with_bbox=True),
+    dict(
+        type='PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor'))
+]
+
+train_dataloader = dict(
+    batch_size=8,
+    num_workers=4,
+    persistent_workers=True,
+    sampler=dict(type='DefaultSampler', shuffle=True),
+    dataset=train_dataset)
+val_dataloader = dict(
+    batch_size=8,
+    num_workers=4,
+    persistent_workers=True,
+    drop_last=False,
+    sampler=dict(type='DefaultSampler', shuffle=False),
+    dataset=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file='annotations/instances_val2017.json',
+        data_prefix=dict(img='val2017/'),
+        test_mode=True,
+        pipeline=test_pipeline,
+        backend_args=backend_args))
+test_dataloader = val_dataloader
+
+val_evaluator = dict(
+    type='CocoMetric',
+    ann_file=data_root + 'annotations/instances_val2017.json',
+    metric='bbox',
+    backend_args=backend_args)
+test_evaluator = val_evaluator
+
+# training settings
+max_epochs = 300
+num_last_epochs = 15
+interval = 10
+
+train_cfg = dict(max_epochs=max_epochs, val_interval=interval)
+
+# optimizer
+# default 8 gpu
+base_lr = 0.01
+optim_wrapper = dict(
+    type='OptimWrapper',
+    optimizer=dict(
+        type='SGD', lr=base_lr, momentum=0.9, weight_decay=5e-4,
+        nesterov=True),
+    paramwise_cfg=dict(norm_decay_mult=0., bias_decay_mult=0.))
+
+# learning rate
+param_scheduler = [
+    dict(
+        # use quadratic formula to warm up 5 epochs
+        # and lr is updated by iteration
+        # TODO: fix default scope in get function
+        type='mmdet.QuadraticWarmupLR',
+        by_epoch=True,
+        begin=0,
+        end=5,
+        convert_to_iter_based=True),
+    dict(
+        # use cosine lr from 5 to 285 epoch
+        type='CosineAnnealingLR',
+        eta_min=base_lr * 0.05,
+        begin=5,
+        T_max=max_epochs - num_last_epochs,
+        end=max_epochs - num_last_epochs,
+        by_epoch=True,
+        convert_to_iter_based=True),
+    dict(
+        # use fixed lr during last 15 epochs
+        type='ConstantLR',
+        by_epoch=True,
+        factor=1,
+        begin=max_epochs - num_last_epochs,
+        end=max_epochs,
+    )
+]
+
+default_hooks = dict(
+    checkpoint=dict(
+        interval=interval,
+        max_keep_ckpts=3  # only keep latest 3 checkpoints
+    ))
+
+custom_hooks = [
+    dict(
+        type='YOLOXModeSwitchHook',
+        num_last_epochs=num_last_epochs,
+        priority=48),
+    dict(type='SyncNormHook', priority=48),
+    dict(
+        type='EMAHook',
+        ema_type='ExpMomentumEMA',
+        momentum=0.0001,
+        update_buffers=True,
+        priority=49)
+]
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# USER SHOULD NOT CHANGE ITS VALUES.
+# base_batch_size = (8 GPUs) x (8 samples per GPU)
+auto_scale_lr = dict(base_batch_size=64)

negi/repos/controlnet_aux/src/controlnet_aux/hed/__init__.py

@@ -0,0 +1,129 @@
+# This is an improved version and model of HED edge detection with Apache License, Version 2.0.
+# Please use this implementation in your products
+# This implementation may produce slightly different results from Saining Xie's official implementations,
+# but it generates smoother edges and is more suitable for ControlNet as well as other image-to-image translations.
+# Different from official models and other implementations, this is an RGB-input model (rather than BGR)
+# and in this way it works better for gradio's RGB protocol
+
+import os
+import warnings
+
+import cv2
+import numpy as np
+import torch
+from einops import rearrange
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from ..util import HWC3, nms, resize_image, safe_step
+
+
+class DoubleConvBlock(torch.nn.Module):
+    def __init__(self, input_channel, output_channel, layer_number):
+        super().__init__()
+        self.convs = torch.nn.Sequential()
+        self.convs.append(torch.nn.Conv2d(in_channels=input_channel, out_channels=output_channel, kernel_size=(3, 3), stride=(1, 1), padding=1))
+        for i in range(1, layer_number):
+            self.convs.append(torch.nn.Conv2d(in_channels=output_channel, out_channels=output_channel, kernel_size=(3, 3), stride=(1, 1), padding=1))
+        self.projection = torch.nn.Conv2d(in_channels=output_channel, out_channels=1, kernel_size=(1, 1), stride=(1, 1), padding=0)
+
+    def __call__(self, x, down_sampling=False):
+        h = x
+        if down_sampling:
+            h = torch.nn.functional.max_pool2d(h, kernel_size=(2, 2), stride=(2, 2))
+        for conv in self.convs:
+            h = conv(h)
+            h = torch.nn.functional.relu(h)
+        return h, self.projection(h)
+
+
+class ControlNetHED_Apache2(torch.nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.norm = torch.nn.Parameter(torch.zeros(size=(1, 3, 1, 1)))
+        self.block1 = DoubleConvBlock(input_channel=3, output_channel=64, layer_number=2)
+        self.block2 = DoubleConvBlock(input_channel=64, output_channel=128, layer_number=2)
+        self.block3 = DoubleConvBlock(input_channel=128, output_channel=256, layer_number=3)
+        self.block4 = DoubleConvBlock(input_channel=256, output_channel=512, layer_number=3)
+        self.block5 = DoubleConvBlock(input_channel=512, output_channel=512, layer_number=3)
+
+    def __call__(self, x):
+        h = x - self.norm
+        h, projection1 = self.block1(h)
+        h, projection2 = self.block2(h, down_sampling=True)
+        h, projection3 = self.block3(h, down_sampling=True)
+        h, projection4 = self.block4(h, down_sampling=True)
+        h, projection5 = self.block5(h, down_sampling=True)
+        return projection1, projection2, projection3, projection4, projection5
+
+class HEDdetector:
+    def __init__(self, netNetwork):
+        self.netNetwork = netNetwork
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_or_path, filename=None, cache_dir=None):
+        filename = filename or "ControlNetHED.pth"
+
+        if os.path.isdir(pretrained_model_or_path):
+            model_path = os.path.join(pretrained_model_or_path, filename)
+        else:
+            model_path = hf_hub_download(pretrained_model_or_path, filename, cache_dir=cache_dir)
+
+        netNetwork = ControlNetHED_Apache2()
+        netNetwork.load_state_dict(torch.load(model_path, map_location='cpu'))
+        netNetwork.float().eval()
+
+        return cls(netNetwork)
+    
+    def to(self, device):
+        self.netNetwork.to(device)
+        return self
+    
+    def __call__(self, input_image, detect_resolution=512, image_resolution=512, safe=False, output_type="pil", scribble=False, **kwargs):
+        if "return_pil" in kwargs:
+            warnings.warn("return_pil is deprecated. Use output_type instead.", DeprecationWarning)
+            output_type = "pil" if kwargs["return_pil"] else "np"
+        if type(output_type) is bool:
+            warnings.warn("Passing `True` or `False` to `output_type` is deprecated and will raise an error in future versions")
+            if output_type:
+                output_type = "pil"
+
+        device = next(iter(self.netNetwork.parameters())).device
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+
+        assert input_image.ndim == 3
+        H, W, C = input_image.shape
+        with torch.no_grad():
+            image_hed = torch.from_numpy(input_image.copy()).float().to(device)
+            image_hed = rearrange(image_hed, 'h w c -> 1 c h w')
+            edges = self.netNetwork(image_hed)
+            edges = [e.detach().cpu().numpy().astype(np.float32)[0, 0] for e in edges]
+            edges = [cv2.resize(e, (W, H), interpolation=cv2.INTER_LINEAR) for e in edges]
+            edges = np.stack(edges, axis=2)
+            edge = 1 / (1 + np.exp(-np.mean(edges, axis=2).astype(np.float64)))
+            if safe:
+                edge = safe_step(edge)
+            edge = (edge * 255.0).clip(0, 255).astype(np.uint8)
+
+        detected_map = edge
+        detected_map = HWC3(detected_map)
+
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+        
+        if scribble:
+            detected_map = nms(detected_map, 127, 3.0)
+            detected_map = cv2.GaussianBlur(detected_map, (0, 0), 3.0)
+            detected_map[detected_map > 4] = 255
+            detected_map[detected_map < 255] = 0
+
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+
+        return detected_map

negi/repos/controlnet_aux/src/controlnet_aux/leres/__init__.py

@@ -0,0 +1,118 @@
+import os
+
+import cv2
+import numpy as np
+import torch
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from ..util import HWC3, resize_image
+from .leres.depthmap import estimateboost, estimateleres
+from .leres.multi_depth_model_woauxi import RelDepthModel
+from .leres.net_tools import strip_prefix_if_present
+from .pix2pix.models.pix2pix4depth_model import Pix2Pix4DepthModel
+from .pix2pix.options.test_options import TestOptions
+
+
+class LeresDetector:
+    def __init__(self, model, pix2pixmodel):
+        self.model = model
+        self.pix2pixmodel = pix2pixmodel
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_or_path, filename=None, pix2pix_filename=None, cache_dir=None):
+        filename = filename or "res101.pth"
+        pix2pix_filename = pix2pix_filename or "latest_net_G.pth"
+
+        if os.path.isdir(pretrained_model_or_path):
+            model_path = os.path.join(pretrained_model_or_path, filename)
+        else:
+            model_path = hf_hub_download(pretrained_model_or_path, filename, cache_dir=cache_dir)
+            
+        checkpoint = torch.load(model_path, map_location=torch.device('cpu'))
+
+        model = RelDepthModel(backbone='resnext101')
+        model.load_state_dict(strip_prefix_if_present(checkpoint['depth_model'], "module."), strict=True)
+        del checkpoint
+
+        if os.path.isdir(pretrained_model_or_path):
+            model_path = os.path.join(pretrained_model_or_path, pix2pix_filename)
+        else:
+            model_path = hf_hub_download(pretrained_model_or_path, pix2pix_filename, cache_dir=cache_dir)
+
+        opt = TestOptions().parse()
+        if not torch.cuda.is_available():
+            opt.gpu_ids = []  # cpu mode
+        pix2pixmodel = Pix2Pix4DepthModel(opt)
+        pix2pixmodel.save_dir = os.path.dirname(model_path)
+        pix2pixmodel.load_networks('latest')
+        pix2pixmodel.eval()
+
+        return cls(model, pix2pixmodel)
+
+    def to(self, device):
+        self.model.to(device)
+        # TODO - refactor pix2pix implementation to support device migration
+        # self.pix2pixmodel.to(device)
+        return self
+
+    def __call__(self, input_image, thr_a=0, thr_b=0, boost=False, detect_resolution=512, image_resolution=512, output_type="pil"):
+        device = next(iter(self.model.parameters())).device
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+        
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+
+        assert input_image.ndim == 3
+        height, width, dim = input_image.shape
+
+        with torch.no_grad():
+
+            if boost:
+                depth = estimateboost(input_image, self.model, 0, self.pix2pixmodel, max(width, height))
+            else:
+                depth = estimateleres(input_image, self.model, width, height)
+
+            numbytes=2
+            depth_min = depth.min()
+            depth_max = depth.max()
+            max_val = (2**(8*numbytes))-1
+
+            # check output before normalizing and mapping to 16 bit
+            if depth_max - depth_min > np.finfo("float").eps:
+                out = max_val * (depth - depth_min) / (depth_max - depth_min)
+            else:
+                out = np.zeros(depth.shape)
+            
+            # single channel, 16 bit image
+            depth_image = out.astype("uint16")
+
+            # convert to uint8
+            depth_image = cv2.convertScaleAbs(depth_image, alpha=(255.0/65535.0))
+
+            # remove near
+            if thr_a != 0:
+                thr_a = ((thr_a/100)*255) 
+                depth_image = cv2.threshold(depth_image, thr_a, 255, cv2.THRESH_TOZERO)[1]
+
+            # invert image
+            depth_image = cv2.bitwise_not(depth_image)
+
+            # remove bg
+            if thr_b != 0:
+                thr_b = ((thr_b/100)*255)
+                depth_image = cv2.threshold(depth_image, thr_b, 255, cv2.THRESH_TOZERO)[1]
+
+        detected_map = depth_image
+        detected_map = HWC3(detected_map)      
+
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+        
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+            
+        return detected_map

negi/repos/controlnet_aux/src/controlnet_aux/leres/leres/LICENSE

@@ -0,0 +1,23 @@
+https://github.com/thygate/stable-diffusion-webui-depthmap-script
+
+MIT License
+
+Copyright (c) 2023 Bob Thiry
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

negi/repos/controlnet_aux/src/controlnet_aux/leres/leres/Resnet.py

@@ -0,0 +1,199 @@
+import torch.nn as nn
+import torch.nn as NN
+
+__all__ = ['ResNet', 'resnet18', 'resnet34', 'resnet50', 'resnet101',
+           'resnet152']
+
+
+model_urls = {
+    'resnet18': 'https://download.pytorch.org/models/resnet18-5c106cde.pth',
+    'resnet34': 'https://download.pytorch.org/models/resnet34-333f7ec4.pth',
+    'resnet50': 'https://download.pytorch.org/models/resnet50-19c8e357.pth',
+    'resnet101': 'https://download.pytorch.org/models/resnet101-5d3b4d8f.pth',
+    'resnet152': 'https://download.pytorch.org/models/resnet152-b121ed2d.pth',
+}
+
+
+def conv3x3(in_planes, out_planes, stride=1):
+    """3x3 convolution with padding"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
+                     padding=1, bias=False)
+
+
+class BasicBlock(nn.Module):
+    expansion = 1
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None):
+        super(BasicBlock, self).__init__()
+        self.conv1 = conv3x3(inplanes, planes, stride)
+        self.bn1 = NN.BatchNorm2d(planes) #NN.BatchNorm2d
+        self.relu = nn.ReLU(inplace=True)
+        self.conv2 = conv3x3(planes, planes)
+        self.bn2 = NN.BatchNorm2d(planes) #NN.BatchNorm2d
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class Bottleneck(nn.Module):
+    expansion = 4
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None):
+        super(Bottleneck, self).__init__()
+        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
+        self.bn1 = NN.BatchNorm2d(planes) #NN.BatchNorm2d
+        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
+                               padding=1, bias=False)
+        self.bn2 = NN.BatchNorm2d(planes) #NN.BatchNorm2d
+        self.conv3 = nn.Conv2d(planes, planes * self.expansion, kernel_size=1, bias=False)
+        self.bn3 = NN.BatchNorm2d(planes * self.expansion) #NN.BatchNorm2d
+        self.relu = nn.ReLU(inplace=True)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+        out = self.relu(out)
+
+        out = self.conv3(out)
+        out = self.bn3(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet(nn.Module):
+
+    def __init__(self, block, layers, num_classes=1000):
+        self.inplanes = 64
+        super(ResNet, self).__init__()
+        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
+                               bias=False)
+        self.bn1 = NN.BatchNorm2d(64)  #NN.BatchNorm2d
+        self.relu = nn.ReLU(inplace=True)
+        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
+        self.layer1 = self._make_layer(block, 64, layers[0])
+        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
+        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
+        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
+        #self.avgpool = nn.AvgPool2d(7, stride=1)
+        #self.fc = nn.Linear(512 * block.expansion, num_classes)
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, nn.BatchNorm2d):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+    def _make_layer(self, block, planes, blocks, stride=1):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                nn.Conv2d(self.inplanes, planes * block.expansion,
+                          kernel_size=1, stride=stride, bias=False),
+                NN.BatchNorm2d(planes * block.expansion), #NN.BatchNorm2d
+            )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        features = []
+
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+        x = self.maxpool(x)
+
+        x = self.layer1(x)
+        features.append(x)
+        x = self.layer2(x)
+        features.append(x)
+        x = self.layer3(x)
+        features.append(x)
+        x = self.layer4(x)
+        features.append(x)
+
+        return features
+
+
+def resnet18(pretrained=True, **kwargs):
+    """Constructs a ResNet-18 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs)
+    return model
+
+
+def resnet34(pretrained=True, **kwargs):
+    """Constructs a ResNet-34 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
+    return model
+
+
+def resnet50(pretrained=True, **kwargs):
+    """Constructs a ResNet-50 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
+
+    return model
+
+
+def resnet101(pretrained=True, **kwargs):
+    """Constructs a ResNet-101 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(Bottleneck, [3, 4, 23, 3], **kwargs)
+
+    return model
+
+
+def resnet152(pretrained=True, **kwargs):
+    """Constructs a ResNet-152 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
+    return model

negi/repos/controlnet_aux/src/controlnet_aux/leres/leres/Resnext_torch.py

@@ -0,0 +1,237 @@
+#!/usr/bin/env python
+# coding: utf-8
+import torch.nn as nn
+
+try:
+    from urllib import urlretrieve
+except ImportError:
+    from urllib.request import urlretrieve
+
+__all__ = ['resnext101_32x8d']
+
+
+model_urls = {
+    'resnext50_32x4d': 'https://download.pytorch.org/models/resnext50_32x4d-7cdf4587.pth',
+    'resnext101_32x8d': 'https://download.pytorch.org/models/resnext101_32x8d-8ba56ff5.pth',
+}
+
+
+def conv3x3(in_planes, out_planes, stride=1, groups=1, dilation=1):
+    """3x3 convolution with padding"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
+                     padding=dilation, groups=groups, bias=False, dilation=dilation)
+
+
+def conv1x1(in_planes, out_planes, stride=1):
+    """1x1 convolution"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)
+
+
+class BasicBlock(nn.Module):
+    expansion = 1
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
+                 base_width=64, dilation=1, norm_layer=None):
+        super(BasicBlock, self).__init__()
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        if groups != 1 or base_width != 64:
+            raise ValueError('BasicBlock only supports groups=1 and base_width=64')
+        if dilation > 1:
+            raise NotImplementedError("Dilation > 1 not supported in BasicBlock")
+        # Both self.conv1 and self.downsample layers downsample the input when stride != 1
+        self.conv1 = conv3x3(inplanes, planes, stride)
+        self.bn1 = norm_layer(planes)
+        self.relu = nn.ReLU(inplace=True)
+        self.conv2 = conv3x3(planes, planes)
+        self.bn2 = norm_layer(planes)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        identity = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+
+        out += identity
+        out = self.relu(out)
+
+        return out
+
+
+class Bottleneck(nn.Module):
+    # Bottleneck in torchvision places the stride for downsampling at 3x3 convolution(self.conv2)
+    # while original implementation places the stride at the first 1x1 convolution(self.conv1)
+    # according to "Deep residual learning for image recognition"https://arxiv.org/abs/1512.03385.
+    # This variant is also known as ResNet V1.5 and improves accuracy according to
+    # https://ngc.nvidia.com/catalog/model-scripts/nvidia:resnet_50_v1_5_for_pytorch.
+
+    expansion = 4
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
+                 base_width=64, dilation=1, norm_layer=None):
+        super(Bottleneck, self).__init__()
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        width = int(planes * (base_width / 64.)) * groups
+        # Both self.conv2 and self.downsample layers downsample the input when stride != 1
+        self.conv1 = conv1x1(inplanes, width)
+        self.bn1 = norm_layer(width)
+        self.conv2 = conv3x3(width, width, stride, groups, dilation)
+        self.bn2 = norm_layer(width)
+        self.conv3 = conv1x1(width, planes * self.expansion)
+        self.bn3 = norm_layer(planes * self.expansion)
+        self.relu = nn.ReLU(inplace=True)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        identity = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+        out = self.relu(out)
+
+        out = self.conv3(out)
+        out = self.bn3(out)
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+
+        out += identity
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet(nn.Module):
+
+    def __init__(self, block, layers, num_classes=1000, zero_init_residual=False,
+                 groups=1, width_per_group=64, replace_stride_with_dilation=None,
+                 norm_layer=None):
+        super(ResNet, self).__init__()
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        self._norm_layer = norm_layer
+
+        self.inplanes = 64
+        self.dilation = 1
+        if replace_stride_with_dilation is None:
+            # each element in the tuple indicates if we should replace
+            # the 2x2 stride with a dilated convolution instead
+            replace_stride_with_dilation = [False, False, False]
+        if len(replace_stride_with_dilation) != 3:
+            raise ValueError("replace_stride_with_dilation should be None "
+                             "or a 3-element tuple, got {}".format(replace_stride_with_dilation))
+        self.groups = groups
+        self.base_width = width_per_group
+        self.conv1 = nn.Conv2d(3, self.inplanes, kernel_size=7, stride=2, padding=3,
+                               bias=False)
+        self.bn1 = norm_layer(self.inplanes)
+        self.relu = nn.ReLU(inplace=True)
+        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
+        self.layer1 = self._make_layer(block, 64, layers[0])
+        self.layer2 = self._make_layer(block, 128, layers[1], stride=2,
+                                       dilate=replace_stride_with_dilation[0])
+        self.layer3 = self._make_layer(block, 256, layers[2], stride=2,
+                                       dilate=replace_stride_with_dilation[1])
+        self.layer4 = self._make_layer(block, 512, layers[3], stride=2,
+                                       dilate=replace_stride_with_dilation[2])
+        #self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
+        #self.fc = nn.Linear(512 * block.expansion, num_classes)
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+        # Zero-initialize the last BN in each residual branch,
+        # so that the residual branch starts with zeros, and each residual block behaves like an identity.
+        # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
+        if zero_init_residual:
+            for m in self.modules():
+                if isinstance(m, Bottleneck):
+                    nn.init.constant_(m.bn3.weight, 0)
+                elif isinstance(m, BasicBlock):
+                    nn.init.constant_(m.bn2.weight, 0)
+
+    def _make_layer(self, block, planes, blocks, stride=1, dilate=False):
+        norm_layer = self._norm_layer
+        downsample = None
+        previous_dilation = self.dilation
+        if dilate:
+            self.dilation *= stride
+            stride = 1
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                conv1x1(self.inplanes, planes * block.expansion, stride),
+                norm_layer(planes * block.expansion),
+            )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample, self.groups,
+                            self.base_width, previous_dilation, norm_layer))
+        self.inplanes = planes * block.expansion
+        for _ in range(1, blocks):
+            layers.append(block(self.inplanes, planes, groups=self.groups,
+                                base_width=self.base_width, dilation=self.dilation,
+                                norm_layer=norm_layer))
+
+        return nn.Sequential(*layers)
+
+    def _forward_impl(self, x):
+        # See note [TorchScript super()]
+        features = []
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+        x = self.maxpool(x)
+
+        x = self.layer1(x)
+        features.append(x)
+
+        x = self.layer2(x)
+        features.append(x)
+
+        x = self.layer3(x)
+        features.append(x)
+
+        x = self.layer4(x)
+        features.append(x)
+
+        #x = self.avgpool(x)
+        #x = torch.flatten(x, 1)
+        #x = self.fc(x)
+
+        return features
+
+    def forward(self, x):
+        return self._forward_impl(x)
+
+
+
+def resnext101_32x8d(pretrained=True, **kwargs):
+    """Constructs a ResNet-152 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    kwargs['groups'] = 32
+    kwargs['width_per_group'] = 8
+
+    model = ResNet(Bottleneck, [3, 4, 23, 3], **kwargs)
+    return model
+

negi/repos/controlnet_aux/src/controlnet_aux/leres/leres/depthmap.py

@@ -0,0 +1,548 @@
+# Author: thygate
+# https://github.com/thygate/stable-diffusion-webui-depthmap-script
+
+import gc
+from operator import getitem
+
+import cv2
+import numpy as np
+import skimage.measure
+import torch
+from torchvision.transforms import transforms
+
+from ...util import torch_gc
+
+whole_size_threshold = 1600  # R_max from the paper
+pix2pixsize = 1024
+
+def scale_torch(img):
+    """
+    Scale the image and output it in torch.tensor.
+    :param img: input rgb is in shape [H, W, C], input depth/disp is in shape [H, W]
+    :param scale: the scale factor. float
+    :return: img. [C, H, W]
+    """
+    if len(img.shape) == 2:
+        img = img[np.newaxis, :, :]
+    if img.shape[2] == 3:
+        transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.485, 0.456, 0.406) , (0.229, 0.224, 0.225) )])
+        img = transform(img.astype(np.float32))
+    else:
+        img = img.astype(np.float32)
+        img = torch.from_numpy(img)
+    return img
+    
+def estimateleres(img, model, w, h):
+    device = next(iter(model.parameters())).device
+    # leres transform input
+    rgb_c = img[:, :, ::-1].copy()
+    A_resize = cv2.resize(rgb_c, (w, h))
+    img_torch = scale_torch(A_resize)[None, :, :, :] 
+    
+    # compute
+    with torch.no_grad():
+        img_torch = img_torch.to(device)
+        prediction = model.depth_model(img_torch)
+
+    prediction = prediction.squeeze().cpu().numpy()
+    prediction = cv2.resize(prediction, (img.shape[1], img.shape[0]), interpolation=cv2.INTER_CUBIC)
+
+    return prediction
+
+def generatemask(size):
+    # Generates a Guassian mask
+    mask = np.zeros(size, dtype=np.float32)
+    sigma = int(size[0]/16)
+    k_size = int(2 * np.ceil(2 * int(size[0]/16)) + 1)
+    mask[int(0.15*size[0]):size[0] - int(0.15*size[0]), int(0.15*size[1]): size[1] - int(0.15*size[1])] = 1
+    mask = cv2.GaussianBlur(mask, (int(k_size), int(k_size)), sigma)
+    mask = (mask - mask.min()) / (mask.max() - mask.min())
+    mask = mask.astype(np.float32)
+    return mask
+
+def resizewithpool(img, size):
+    i_size = img.shape[0]
+    n = int(np.floor(i_size/size))
+
+    out = skimage.measure.block_reduce(img, (n, n), np.max)
+    return out
+
+def rgb2gray(rgb):
+    # Converts rgb to gray
+    return np.dot(rgb[..., :3], [0.2989, 0.5870, 0.1140])
+
+def calculateprocessingres(img, basesize, confidence=0.1, scale_threshold=3, whole_size_threshold=3000):
+    # Returns the R_x resolution described in section 5 of the main paper.
+
+    # Parameters:
+    #    img :input rgb image
+    #    basesize : size the dilation kernel which is equal to receptive field of the network.
+    #    confidence: value of x in R_x; allowed percentage of pixels that are not getting any contextual cue.
+    #    scale_threshold: maximum allowed upscaling on the input image ; it has been set to 3.
+    #    whole_size_threshold: maximum allowed resolution. (R_max from section 6 of the main paper)
+
+    # Returns:
+    #    outputsize_scale*speed_scale :The computed R_x resolution
+    #    patch_scale: K parameter from section 6 of the paper
+
+    # speed scale parameter is to process every image in a smaller size to accelerate the R_x resolution search
+    speed_scale = 32
+    image_dim = int(min(img.shape[0:2]))
+
+    gray = rgb2gray(img)
+    grad = np.abs(cv2.Sobel(gray, cv2.CV_64F, 0, 1, ksize=3)) + np.abs(cv2.Sobel(gray, cv2.CV_64F, 1, 0, ksize=3))
+    grad = cv2.resize(grad, (image_dim, image_dim), cv2.INTER_AREA)
+
+    # thresholding the gradient map to generate the edge-map as a proxy of the contextual cues
+    m = grad.min()
+    M = grad.max()
+    middle = m + (0.4 * (M - m))
+    grad[grad < middle] = 0
+    grad[grad >= middle] = 1
+
+    # dilation kernel with size of the receptive field
+    kernel = np.ones((int(basesize/speed_scale), int(basesize/speed_scale)), float)
+    # dilation kernel with size of the a quarter of receptive field used to compute k
+    # as described in section 6 of main paper
+    kernel2 = np.ones((int(basesize / (4*speed_scale)), int(basesize / (4*speed_scale))), float)
+
+    # Output resolution limit set by the whole_size_threshold and scale_threshold.
+    threshold = min(whole_size_threshold, scale_threshold * max(img.shape[:2]))
+
+    outputsize_scale = basesize / speed_scale
+    for p_size in range(int(basesize/speed_scale), int(threshold/speed_scale), int(basesize / (2*speed_scale))):
+        grad_resized = resizewithpool(grad, p_size)
+        grad_resized = cv2.resize(grad_resized, (p_size, p_size), cv2.INTER_NEAREST)
+        grad_resized[grad_resized >= 0.5] = 1
+        grad_resized[grad_resized < 0.5] = 0
+
+        dilated = cv2.dilate(grad_resized, kernel, iterations=1)
+        meanvalue = (1-dilated).mean()
+        if meanvalue > confidence:
+            break
+        else:
+            outputsize_scale = p_size
+
+    grad_region = cv2.dilate(grad_resized, kernel2, iterations=1)
+    patch_scale = grad_region.mean()
+
+    return int(outputsize_scale*speed_scale), patch_scale
+
+# Generate a double-input depth estimation
+def doubleestimate(img, size1, size2, pix2pixsize, model, net_type, pix2pixmodel):
+    # Generate the low resolution estimation
+    estimate1 = singleestimate(img, size1, model, net_type)
+    # Resize to the inference size of merge network.
+    estimate1 = cv2.resize(estimate1, (pix2pixsize, pix2pixsize), interpolation=cv2.INTER_CUBIC)
+
+    # Generate the high resolution estimation
+    estimate2 = singleestimate(img, size2, model, net_type)
+    # Resize to the inference size of merge network.
+    estimate2 = cv2.resize(estimate2, (pix2pixsize, pix2pixsize), interpolation=cv2.INTER_CUBIC)
+
+    # Inference on the merge model
+    pix2pixmodel.set_input(estimate1, estimate2)
+    pix2pixmodel.test()
+    visuals = pix2pixmodel.get_current_visuals()
+    prediction_mapped = visuals['fake_B']
+    prediction_mapped = (prediction_mapped+1)/2
+    prediction_mapped = (prediction_mapped - torch.min(prediction_mapped)) / (
+                torch.max(prediction_mapped) - torch.min(prediction_mapped))
+    prediction_mapped = prediction_mapped.squeeze().cpu().numpy()
+
+    return prediction_mapped
+
+# Generate a single-input depth estimation
+def singleestimate(img, msize, model, net_type):
+    # if net_type == 0:
+    return estimateleres(img, model, msize, msize)
+    # else:
+    # 	return estimatemidasBoost(img, model, msize, msize)
+
+def applyGridpatch(blsize, stride, img, box):
+    # Extract a simple grid patch.
+    counter1 = 0
+    patch_bound_list = {}
+    for k in range(blsize, img.shape[1] - blsize, stride):
+        for j in range(blsize, img.shape[0] - blsize, stride):
+            patch_bound_list[str(counter1)] = {}
+            patchbounds = [j - blsize, k - blsize, j - blsize + 2 * blsize, k - blsize + 2 * blsize]
+            patch_bound = [box[0] + patchbounds[1], box[1] + patchbounds[0], patchbounds[3] - patchbounds[1],
+                           patchbounds[2] - patchbounds[0]]
+            patch_bound_list[str(counter1)]['rect'] = patch_bound
+            patch_bound_list[str(counter1)]['size'] = patch_bound[2]
+            counter1 = counter1 + 1
+    return patch_bound_list
+
+# Generating local patches to perform the local refinement described in section 6 of the main paper.
+def generatepatchs(img, base_size):
+    
+    # Compute the gradients as a proxy of the contextual cues.
+    img_gray = rgb2gray(img)
+    whole_grad = np.abs(cv2.Sobel(img_gray, cv2.CV_64F, 0, 1, ksize=3)) +\
+        np.abs(cv2.Sobel(img_gray, cv2.CV_64F, 1, 0, ksize=3))
+
+    threshold = whole_grad[whole_grad > 0].mean()
+    whole_grad[whole_grad < threshold] = 0
+
+    # We use the integral image to speed-up the evaluation of the amount of gradients for each patch.
+    gf = whole_grad.sum()/len(whole_grad.reshape(-1))
+    grad_integral_image = cv2.integral(whole_grad)
+
+    # Variables are selected such that the initial patch size would be the receptive field size
+    # and the stride is set to 1/3 of the receptive field size.
+    blsize = int(round(base_size/2))
+    stride = int(round(blsize*0.75))
+
+    # Get initial Grid
+    patch_bound_list = applyGridpatch(blsize, stride, img, [0, 0, 0, 0])
+
+    # Refine initial Grid of patches by discarding the flat (in terms of gradients of the rgb image) ones. Refine
+    # each patch size to ensure that there will be enough depth cues for the network to generate a consistent depth map.
+    print("Selecting patches ...")
+    patch_bound_list = adaptiveselection(grad_integral_image, patch_bound_list, gf)
+
+    # Sort the patch list to make sure the merging operation will be done with the correct order: starting from biggest
+    # patch
+    patchset = sorted(patch_bound_list.items(), key=lambda x: getitem(x[1], 'size'), reverse=True)
+    return patchset
+
+def getGF_fromintegral(integralimage, rect):
+    # Computes the gradient density of a given patch from the gradient integral image.
+    x1 = rect[1]
+    x2 = rect[1]+rect[3]
+    y1 = rect[0]
+    y2 = rect[0]+rect[2]
+    value = integralimage[x2, y2]-integralimage[x1, y2]-integralimage[x2, y1]+integralimage[x1, y1]
+    return value
+
+# Adaptively select patches
+def adaptiveselection(integral_grad, patch_bound_list, gf):
+    patchlist = {}
+    count = 0
+    height, width = integral_grad.shape
+
+    search_step = int(32/factor)
+
+    # Go through all patches
+    for c in range(len(patch_bound_list)):
+        # Get patch
+        bbox = patch_bound_list[str(c)]['rect']
+
+        # Compute the amount of gradients present in the patch from the integral image.
+        cgf = getGF_fromintegral(integral_grad, bbox)/(bbox[2]*bbox[3])
+
+        # Check if patching is beneficial by comparing the gradient density of the patch to
+        # the gradient density of the whole image
+        if cgf >= gf:
+            bbox_test = bbox.copy()
+            patchlist[str(count)] = {}
+
+            # Enlarge each patch until the gradient density of the patch is equal
+            # to the whole image gradient density
+            while True:
+
+                bbox_test[0] = bbox_test[0] - int(search_step/2)
+                bbox_test[1] = bbox_test[1] - int(search_step/2)
+
+                bbox_test[2] = bbox_test[2] + search_step
+                bbox_test[3] = bbox_test[3] + search_step
+
+                # Check if we are still within the image
+                if bbox_test[0] < 0 or bbox_test[1] < 0 or bbox_test[1] + bbox_test[3] >= height \
+                        or bbox_test[0] + bbox_test[2] >= width:
+                    break
+
+                # Compare gradient density
+                cgf = getGF_fromintegral(integral_grad, bbox_test)/(bbox_test[2]*bbox_test[3])
+                if cgf < gf:
+                    break
+                bbox = bbox_test.copy()
+
+            # Add patch to selected patches
+            patchlist[str(count)]['rect'] = bbox
+            patchlist[str(count)]['size'] = bbox[2]
+            count = count + 1
+    
+    # Return selected patches
+    return patchlist
+
+def impatch(image, rect):
+    # Extract the given patch pixels from a given image.
+    w1 = rect[0]
+    h1 = rect[1]
+    w2 = w1 + rect[2]
+    h2 = h1 + rect[3]
+    image_patch = image[h1:h2, w1:w2]
+    return image_patch
+
+class ImageandPatchs:
+    def __init__(self, root_dir, name, patchsinfo, rgb_image, scale=1):
+        self.root_dir = root_dir
+        self.patchsinfo = patchsinfo
+        self.name = name
+        self.patchs = patchsinfo
+        self.scale = scale
+
+        self.rgb_image = cv2.resize(rgb_image, (round(rgb_image.shape[1]*scale), round(rgb_image.shape[0]*scale)),
+                                    interpolation=cv2.INTER_CUBIC)
+
+        self.do_have_estimate = False
+        self.estimation_updated_image = None
+        self.estimation_base_image = None
+
+    def __len__(self):
+        return len(self.patchs)
+
+    def set_base_estimate(self, est):
+        self.estimation_base_image = est
+        if self.estimation_updated_image is not None:
+            self.do_have_estimate = True
+
+    def set_updated_estimate(self, est):
+        self.estimation_updated_image = est
+        if self.estimation_base_image is not None:
+            self.do_have_estimate = True
+
+    def __getitem__(self, index):
+        patch_id = int(self.patchs[index][0])
+        rect = np.array(self.patchs[index][1]['rect'])
+        msize = self.patchs[index][1]['size']
+
+        ## applying scale to rect:
+        rect = np.round(rect * self.scale)
+        rect = rect.astype('int')
+        msize = round(msize * self.scale)
+
+        patch_rgb = impatch(self.rgb_image, rect)
+        if self.do_have_estimate:
+            patch_whole_estimate_base = impatch(self.estimation_base_image, rect)
+            patch_whole_estimate_updated = impatch(self.estimation_updated_image, rect)
+            return {'patch_rgb': patch_rgb, 'patch_whole_estimate_base': patch_whole_estimate_base,
+                    'patch_whole_estimate_updated': patch_whole_estimate_updated, 'rect': rect,
+                    'size': msize, 'id': patch_id}
+        else:
+            return {'patch_rgb': patch_rgb, 'rect': rect, 'size': msize, 'id': patch_id}
+
+    def print_options(self, opt):
+        """Print and save options
+
+        It will print both current options and default values(if different).
+        It will save options into a text file / [checkpoints_dir] / opt.txt
+        """
+        message = ''
+        message += '----------------- Options ---------------\n'
+        for k, v in sorted(vars(opt).items()):
+            comment = ''
+            default = self.parser.get_default(k)
+            if v != default:
+                comment = '\t[default: %s]' % str(default)
+            message += '{:>25}: {:<30}{}\n'.format(str(k), str(v), comment)
+        message += '----------------- End -------------------'
+        print(message)
+
+        # save to the disk
+        """
+        expr_dir = os.path.join(opt.checkpoints_dir, opt.name)
+        util.mkdirs(expr_dir)
+        file_name = os.path.join(expr_dir, '{}_opt.txt'.format(opt.phase))
+        with open(file_name, 'wt') as opt_file:
+            opt_file.write(message)
+            opt_file.write('\n')
+        """
+
+    def parse(self):
+        """Parse our options, create checkpoints directory suffix, and set up gpu device."""
+        opt = self.gather_options()
+        opt.isTrain = self.isTrain   # train or test
+
+        # process opt.suffix
+        if opt.suffix:
+            suffix = ('_' + opt.suffix.format(**vars(opt))) if opt.suffix != '' else ''
+            opt.name = opt.name + suffix
+
+        #self.print_options(opt)
+
+        # set gpu ids
+        str_ids = opt.gpu_ids.split(',')
+        opt.gpu_ids = []
+        for str_id in str_ids:
+            id = int(str_id)
+            if id >= 0:
+                opt.gpu_ids.append(id)
+        #if len(opt.gpu_ids) > 0:
+        #    torch.cuda.set_device(opt.gpu_ids[0])
+
+        self.opt = opt
+        return self.opt
+
+
+def estimateboost(img, model, model_type, pix2pixmodel, max_res=512, depthmap_script_boost_rmax=None):
+    global whole_size_threshold
+    
+    # get settings
+    if depthmap_script_boost_rmax:
+        whole_size_threshold = depthmap_script_boost_rmax
+        
+    if model_type == 0: #leres
+        net_receptive_field_size = 448
+        patch_netsize = 2 * net_receptive_field_size
+    elif model_type == 1: #dpt_beit_large_512
+        net_receptive_field_size = 512
+        patch_netsize = 2 * net_receptive_field_size
+    else: #other midas
+        net_receptive_field_size = 384
+        patch_netsize = 2 * net_receptive_field_size
+
+    gc.collect()
+    torch_gc()
+
+    # Generate mask used to smoothly blend the local pathc estimations to the base estimate.
+    # It is arbitrarily large to avoid artifacts during rescaling for each crop.
+    mask_org = generatemask((3000, 3000))
+    mask = mask_org.copy()
+
+    # Value x of R_x defined in the section 5 of the main paper.
+    r_threshold_value = 0.2
+    #if R0:
+    #	r_threshold_value = 0
+
+    input_resolution = img.shape
+    scale_threshold = 3  # Allows up-scaling with a scale up to 3
+
+    # Find the best input resolution R-x. The resolution search described in section 5-double estimation of the main paper and section B of the
+    # supplementary material.
+    whole_image_optimal_size, patch_scale = calculateprocessingres(img, net_receptive_field_size, r_threshold_value, scale_threshold, whole_size_threshold)
+
+    # print('wholeImage being processed in :', whole_image_optimal_size)
+
+    # Generate the base estimate using the double estimation.
+    whole_estimate = doubleestimate(img, net_receptive_field_size, whole_image_optimal_size, pix2pixsize, model, model_type, pix2pixmodel)
+
+    # Compute the multiplier described in section 6 of the main paper to make sure our initial patch can select
+    # small high-density regions of the image.
+    global factor
+    factor = max(min(1, 4 * patch_scale * whole_image_optimal_size / whole_size_threshold), 0.2)
+    # print('Adjust factor is:', 1/factor)
+        
+    # Check if Local boosting is beneficial.
+    if max_res < whole_image_optimal_size:
+        # print("No Local boosting. Specified Max Res is smaller than R20, Returning doubleestimate result")
+        return cv2.resize(whole_estimate, (input_resolution[1], input_resolution[0]), interpolation=cv2.INTER_CUBIC)
+
+    # Compute the default target resolution.
+    if img.shape[0] > img.shape[1]:
+        a = 2 * whole_image_optimal_size
+        b = round(2 * whole_image_optimal_size * img.shape[1] / img.shape[0])
+    else:
+        a = round(2 * whole_image_optimal_size * img.shape[0] / img.shape[1])
+        b = 2 * whole_image_optimal_size
+    b = int(round(b / factor))
+    a = int(round(a / factor))
+
+    """
+    # recompute a, b and saturate to max res.
+    if max(a,b) > max_res:
+        print('Default Res is higher than max-res: Reducing final resolution')
+        if img.shape[0] > img.shape[1]:
+            a = max_res
+            b = round(max_res * img.shape[1] / img.shape[0])
+        else:
+            a = round(max_res * img.shape[0] / img.shape[1])
+            b = max_res
+        b = int(b)
+        a = int(a)
+    """
+
+    img = cv2.resize(img, (b, a), interpolation=cv2.INTER_CUBIC)
+
+    # Extract selected patches for local refinement
+    base_size = net_receptive_field_size * 2
+    patchset = generatepatchs(img, base_size)
+
+    # print('Target resolution: ', img.shape)
+
+    # Computing a scale in case user prompted to generate the results as the same resolution of the input.
+    # Notice that our method output resolution is independent of the input resolution and this parameter will only
+    # enable a scaling operation during the local patch merge implementation to generate results with the same resolution
+    # as the input.
+    """
+    if output_resolution == 1:
+        mergein_scale = input_resolution[0] / img.shape[0]
+        print('Dynamicly change merged-in resolution; scale:', mergein_scale)
+    else:
+        mergein_scale = 1
+    """
+    # always rescale to input res for now
+    mergein_scale = input_resolution[0] / img.shape[0]
+
+    imageandpatchs = ImageandPatchs('', '', patchset, img, mergein_scale)
+    whole_estimate_resized = cv2.resize(whole_estimate, (round(img.shape[1]*mergein_scale),
+                                        round(img.shape[0]*mergein_scale)), interpolation=cv2.INTER_CUBIC)
+    imageandpatchs.set_base_estimate(whole_estimate_resized.copy())
+    imageandpatchs.set_updated_estimate(whole_estimate_resized.copy())
+
+    print('Resulting depthmap resolution will be :', whole_estimate_resized.shape[:2])
+    print('Patches to process: '+str(len(imageandpatchs)))
+
+    # Enumerate through all patches, generate their estimations and refining the base estimate.
+    for patch_ind in range(len(imageandpatchs)):
+        
+        # Get patch information
+        patch = imageandpatchs[patch_ind] # patch object
+        patch_rgb = patch['patch_rgb'] # rgb patch
+        patch_whole_estimate_base = patch['patch_whole_estimate_base'] # corresponding patch from base
+        rect = patch['rect'] # patch size and location
+        patch_id = patch['id'] # patch ID
+        org_size = patch_whole_estimate_base.shape # the original size from the unscaled input
+        print('\t Processing patch', patch_ind, '/', len(imageandpatchs)-1, '|', rect)
+
+        # We apply double estimation for patches. The high resolution value is fixed to twice the receptive
+        # field size of the network for patches to accelerate the process.
+        patch_estimation = doubleestimate(patch_rgb, net_receptive_field_size, patch_netsize, pix2pixsize, model, model_type, pix2pixmodel)
+        patch_estimation = cv2.resize(patch_estimation, (pix2pixsize, pix2pixsize), interpolation=cv2.INTER_CUBIC)
+        patch_whole_estimate_base = cv2.resize(patch_whole_estimate_base, (pix2pixsize, pix2pixsize), interpolation=cv2.INTER_CUBIC)
+
+        # Merging the patch estimation into the base estimate using our merge network:
+        # We feed the patch estimation and the same region from the updated base estimate to the merge network
+        # to generate the target estimate for the corresponding region.
+        pix2pixmodel.set_input(patch_whole_estimate_base, patch_estimation)
+
+        # Run merging network
+        pix2pixmodel.test()
+        visuals = pix2pixmodel.get_current_visuals()
+
+        prediction_mapped = visuals['fake_B']
+        prediction_mapped = (prediction_mapped+1)/2
+        prediction_mapped = prediction_mapped.squeeze().cpu().numpy()
+
+        mapped = prediction_mapped
+
+        # We use a simple linear polynomial to make sure the result of the merge network would match the values of
+        # base estimate
+        p_coef = np.polyfit(mapped.reshape(-1), patch_whole_estimate_base.reshape(-1), deg=1)
+        merged = np.polyval(p_coef, mapped.reshape(-1)).reshape(mapped.shape)
+
+        merged = cv2.resize(merged, (org_size[1],org_size[0]), interpolation=cv2.INTER_CUBIC)
+
+        # Get patch size and location
+        w1 = rect[0]
+        h1 = rect[1]
+        w2 = w1 + rect[2]
+        h2 = h1 + rect[3]
+
+        # To speed up the implementation, we only generate the Gaussian mask once with a sufficiently large size
+        # and resize it to our needed size while merging the patches.
+        if mask.shape != org_size:
+            mask = cv2.resize(mask_org, (org_size[1],org_size[0]), interpolation=cv2.INTER_LINEAR)
+
+        tobemergedto = imageandpatchs.estimation_updated_image
+
+        # Update the whole estimation:
+        # We use a simple Gaussian mask to blend the merged patch region with the base estimate to ensure seamless
+        # blending at the boundaries of the patch region.
+        tobemergedto[h1:h2, w1:w2] = np.multiply(tobemergedto[h1:h2, w1:w2], 1 - mask) + np.multiply(merged, mask)
+        imageandpatchs.set_updated_estimate(tobemergedto)
+
+    # output
+    return cv2.resize(imageandpatchs.estimation_updated_image, (input_resolution[1], input_resolution[0]), interpolation=cv2.INTER_CUBIC)

negi/repos/controlnet_aux/src/controlnet_aux/leres/leres/multi_depth_model_woauxi.py

@@ -0,0 +1,35 @@
+import torch
+import torch.nn as nn
+
+from . import network_auxi as network
+from .net_tools import get_func
+
+
+class RelDepthModel(nn.Module):
+    def __init__(self, backbone='resnet50'):
+        super(RelDepthModel, self).__init__()
+        if backbone == 'resnet50':
+            encoder = 'resnet50_stride32'
+        elif backbone == 'resnext101':
+            encoder = 'resnext101_stride32x8d'
+        self.depth_model = DepthModel(encoder)
+
+    def inference(self, rgb):
+        with torch.no_grad():
+            input = rgb.to(self.depth_model.device)
+            depth = self.depth_model(input)
+            #pred_depth_out = depth - depth.min() + 0.01
+            return depth #pred_depth_out
+
+
+class DepthModel(nn.Module):
+    def __init__(self, encoder):
+        super(DepthModel, self).__init__()
+        backbone = network.__name__.split('.')[-1] + '.' + encoder
+        self.encoder_modules = get_func(backbone)()
+        self.decoder_modules = network.Decoder()
+
+    def forward(self, x):
+        lateral_out = self.encoder_modules(x)
+        out_logit = self.decoder_modules(lateral_out)
+        return out_logit

negi/repos/controlnet_aux/src/controlnet_aux/leres/leres/net_tools.py

@@ -0,0 +1,54 @@
+import importlib
+import torch
+import os
+from collections import OrderedDict
+
+
+def get_func(func_name):
+    """Helper to return a function object by name. func_name must identify a
+    function in this module or the path to a function relative to the base
+    'modeling' module.
+    """
+    if func_name == '':
+        return None
+    try:
+        parts = func_name.split('.')
+        # Refers to a function in this module
+        if len(parts) == 1:
+            return globals()[parts[0]]
+        # Otherwise, assume we're referencing a module under modeling
+        module_name = 'controlnet_aux.leres.leres.' + '.'.join(parts[:-1])
+        module = importlib.import_module(module_name)
+        return getattr(module, parts[-1])
+    except Exception:
+        print('Failed to f1ind function: %s', func_name)
+        raise
+
+def load_ckpt(args, depth_model, shift_model, focal_model):
+    """
+    Load checkpoint.
+    """
+    if os.path.isfile(args.load_ckpt):
+        print("loading checkpoint %s" % args.load_ckpt)
+        checkpoint = torch.load(args.load_ckpt)
+        if shift_model is not None:
+            shift_model.load_state_dict(strip_prefix_if_present(checkpoint['shift_model'], 'module.'),
+                                    strict=True)
+        if focal_model is not None:
+            focal_model.load_state_dict(strip_prefix_if_present(checkpoint['focal_model'], 'module.'),
+                                    strict=True)
+        depth_model.load_state_dict(strip_prefix_if_present(checkpoint['depth_model'], "module."),
+                                    strict=True)
+        del checkpoint
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
+
+def strip_prefix_if_present(state_dict, prefix):
+    keys = sorted(state_dict.keys())
+    if not all(key.startswith(prefix) for key in keys):
+        return state_dict
+    stripped_state_dict = OrderedDict()
+    for key, value in state_dict.items():
+        stripped_state_dict[key.replace(prefix, "")] = value
+    return stripped_state_dict

negi/repos/controlnet_aux/src/controlnet_aux/leres/leres/network_auxi.py

@@ -0,0 +1,417 @@
+import torch
+import torch.nn as nn
+import torch.nn.init as init
+
+from . import Resnet, Resnext_torch
+
+
+def resnet50_stride32():
+    return DepthNet(backbone='resnet', depth=50, upfactors=[2, 2, 2, 2])
+
+def resnext101_stride32x8d():
+    return DepthNet(backbone='resnext101_32x8d', depth=101, upfactors=[2, 2, 2, 2])
+
+
+class Decoder(nn.Module):
+    def __init__(self):
+        super(Decoder, self).__init__()
+        self.inchannels =  [256, 512, 1024, 2048]
+        self.midchannels = [256, 256, 256, 512]
+        self.upfactors = [2,2,2,2]
+        self.outchannels = 1
+
+        self.conv = FTB(inchannels=self.inchannels[3], midchannels=self.midchannels[3])
+        self.conv1 = nn.Conv2d(in_channels=self.midchannels[3], out_channels=self.midchannels[2], kernel_size=3, padding=1, stride=1, bias=True)
+        self.upsample = nn.Upsample(scale_factor=self.upfactors[3], mode='bilinear', align_corners=True)
+        
+        self.ffm2 = FFM(inchannels=self.inchannels[2], midchannels=self.midchannels[2], outchannels = self.midchannels[2], upfactor=self.upfactors[2])
+        self.ffm1 = FFM(inchannels=self.inchannels[1], midchannels=self.midchannels[1], outchannels = self.midchannels[1], upfactor=self.upfactors[1])
+        self.ffm0 = FFM(inchannels=self.inchannels[0], midchannels=self.midchannels[0], outchannels = self.midchannels[0], upfactor=self.upfactors[0])
+        
+        self.outconv = AO(inchannels=self.midchannels[0], outchannels=self.outchannels, upfactor=2)
+        self._init_params()
+        
+    def _init_params(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.ConvTranspose2d):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.BatchNorm2d): #NN.BatchNorm2d
+                init.constant_(m.weight, 1)
+                init.constant_(m.bias, 0)
+            elif isinstance(m, nn.Linear):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+                    
+    def forward(self, features):
+        x_32x = self.conv(features[3])  # 1/32
+        x_32 = self.conv1(x_32x)
+        x_16 = self.upsample(x_32)  # 1/16
+
+        x_8 = self.ffm2(features[2], x_16)  # 1/8
+        x_4 = self.ffm1(features[1], x_8)  # 1/4
+        x_2 = self.ffm0(features[0], x_4)  # 1/2
+        #-----------------------------------------
+        x = self.outconv(x_2)  # original size
+        return x
+
+class DepthNet(nn.Module):
+    __factory = {
+        18: Resnet.resnet18,
+        34: Resnet.resnet34,
+        50: Resnet.resnet50,
+        101: Resnet.resnet101,
+        152: Resnet.resnet152
+    }
+    def __init__(self,
+                 backbone='resnet',
+                 depth=50,
+                 upfactors=[2, 2, 2, 2]):
+        super(DepthNet, self).__init__()
+        self.backbone = backbone
+        self.depth = depth
+        self.pretrained = False
+        self.inchannels = [256, 512, 1024, 2048]
+        self.midchannels = [256, 256, 256, 512]
+        self.upfactors = upfactors
+        self.outchannels = 1
+
+        # Build model
+        if self.backbone == 'resnet':
+            if self.depth not in DepthNet.__factory:
+                raise KeyError("Unsupported depth:", self.depth)
+            self.encoder = DepthNet.__factory[depth](pretrained=self.pretrained)
+        elif self.backbone == 'resnext101_32x8d':
+            self.encoder = Resnext_torch.resnext101_32x8d(pretrained=self.pretrained)
+        else:
+            self.encoder = Resnext_torch.resnext101(pretrained=self.pretrained)
+
+    def forward(self, x):
+        x = self.encoder(x)  # 1/32, 1/16, 1/8, 1/4
+        return x
+
+
+class FTB(nn.Module):
+    def __init__(self, inchannels, midchannels=512):
+        super(FTB, self).__init__()
+        self.in1 = inchannels
+        self.mid = midchannels
+        self.conv1 = nn.Conv2d(in_channels=self.in1, out_channels=self.mid, kernel_size=3, padding=1, stride=1,
+                               bias=True)
+        # NN.BatchNorm2d
+        self.conv_branch = nn.Sequential(nn.ReLU(inplace=True), \
+                                         nn.Conv2d(in_channels=self.mid, out_channels=self.mid, kernel_size=3,
+                                                   padding=1, stride=1, bias=True), \
+                                         nn.BatchNorm2d(num_features=self.mid), \
+                                         nn.ReLU(inplace=True), \
+                                         nn.Conv2d(in_channels=self.mid, out_channels=self.mid, kernel_size=3,
+                                                   padding=1, stride=1, bias=True))
+        self.relu = nn.ReLU(inplace=True)
+
+        self.init_params()
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = x + self.conv_branch(x)
+        x = self.relu(x)
+
+        return x
+
+    def init_params(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.ConvTranspose2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                init.normal_(m.weight, std=0.01)
+                # init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.BatchNorm2d):  # NN.BatchNorm2d
+                init.constant_(m.weight, 1)
+                init.constant_(m.bias, 0)
+            elif isinstance(m, nn.Linear):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+
+
+class ATA(nn.Module):
+    def __init__(self, inchannels, reduction=8):
+        super(ATA, self).__init__()
+        self.inchannels = inchannels
+        self.avg_pool = nn.AdaptiveAvgPool2d(1)
+        self.fc = nn.Sequential(nn.Linear(self.inchannels * 2, self.inchannels // reduction),
+                                nn.ReLU(inplace=True),
+                                nn.Linear(self.inchannels // reduction, self.inchannels),
+                                nn.Sigmoid())
+        self.init_params()
+
+    def forward(self, low_x, high_x):
+        n, c, _, _ = low_x.size()
+        x = torch.cat([low_x, high_x], 1)
+        x = self.avg_pool(x)
+        x = x.view(n, -1)
+        x = self.fc(x).view(n, c, 1, 1)
+        x = low_x * x + high_x
+
+        return x
+
+    def init_params(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                # init.normal(m.weight, std=0.01)
+                init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.ConvTranspose2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                # init.normal_(m.weight, std=0.01)
+                init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.BatchNorm2d):  # NN.BatchNorm2d
+                init.constant_(m.weight, 1)
+                init.constant_(m.bias, 0)
+            elif isinstance(m, nn.Linear):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+
+
+class FFM(nn.Module):
+    def __init__(self, inchannels, midchannels, outchannels, upfactor=2):
+        super(FFM, self).__init__()
+        self.inchannels = inchannels
+        self.midchannels = midchannels
+        self.outchannels = outchannels
+        self.upfactor = upfactor
+
+        self.ftb1 = FTB(inchannels=self.inchannels, midchannels=self.midchannels)
+        # self.ata = ATA(inchannels = self.midchannels)
+        self.ftb2 = FTB(inchannels=self.midchannels, midchannels=self.outchannels)
+
+        self.upsample = nn.Upsample(scale_factor=self.upfactor, mode='bilinear', align_corners=True)
+
+        self.init_params()
+
+    def forward(self, low_x, high_x):
+        x = self.ftb1(low_x)
+        x = x + high_x
+        x = self.ftb2(x)
+        x = self.upsample(x)
+
+        return x
+
+    def init_params(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                init.normal_(m.weight, std=0.01)
+                # init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.ConvTranspose2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                init.normal_(m.weight, std=0.01)
+                # init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.BatchNorm2d):  # NN.Batchnorm2d
+                init.constant_(m.weight, 1)
+                init.constant_(m.bias, 0)
+            elif isinstance(m, nn.Linear):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+
+
+class AO(nn.Module):
+    # Adaptive output module
+    def __init__(self, inchannels, outchannels, upfactor=2):
+        super(AO, self).__init__()
+        self.inchannels = inchannels
+        self.outchannels = outchannels
+        self.upfactor = upfactor
+
+        self.adapt_conv = nn.Sequential(
+            nn.Conv2d(in_channels=self.inchannels, out_channels=self.inchannels // 2, kernel_size=3, padding=1,
+                      stride=1, bias=True), \
+            nn.BatchNorm2d(num_features=self.inchannels // 2), \
+            nn.ReLU(inplace=True), \
+            nn.Conv2d(in_channels=self.inchannels // 2, out_channels=self.outchannels, kernel_size=3, padding=1,
+                      stride=1, bias=True), \
+            nn.Upsample(scale_factor=self.upfactor, mode='bilinear', align_corners=True))
+
+        self.init_params()
+
+    def forward(self, x):
+        x = self.adapt_conv(x)
+        return x
+
+    def init_params(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                init.normal_(m.weight, std=0.01)
+                # init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.ConvTranspose2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                init.normal_(m.weight, std=0.01)
+                # init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.BatchNorm2d):  # NN.Batchnorm2d
+                init.constant_(m.weight, 1)
+                init.constant_(m.bias, 0)
+            elif isinstance(m, nn.Linear):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+
+
+
+# ==============================================================================================================
+
+
+class ResidualConv(nn.Module):
+    def __init__(self, inchannels):
+        super(ResidualConv, self).__init__()
+        # NN.BatchNorm2d
+        self.conv = nn.Sequential(
+            # nn.BatchNorm2d(num_features=inchannels),
+            nn.ReLU(inplace=False),
+            # nn.Conv2d(in_channels=inchannels, out_channels=inchannels, kernel_size=3, padding=1, stride=1, groups=inchannels,bias=True),
+            # nn.Conv2d(in_channels=inchannels, out_channels=inchannels, kernel_size=1, padding=0, stride=1, groups=1,bias=True)
+            nn.Conv2d(in_channels=inchannels, out_channels=inchannels / 2, kernel_size=3, padding=1, stride=1,
+                      bias=False),
+            nn.BatchNorm2d(num_features=inchannels / 2),
+            nn.ReLU(inplace=False),
+            nn.Conv2d(in_channels=inchannels / 2, out_channels=inchannels, kernel_size=3, padding=1, stride=1,
+                      bias=False)
+        )
+        self.init_params()
+
+    def forward(self, x):
+        x = self.conv(x) + x
+        return x
+
+    def init_params(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                init.normal_(m.weight, std=0.01)
+                # init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.ConvTranspose2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                init.normal_(m.weight, std=0.01)
+                # init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.BatchNorm2d):  # NN.BatchNorm2d
+                init.constant_(m.weight, 1)
+                init.constant_(m.bias, 0)
+            elif isinstance(m, nn.Linear):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+
+
+class FeatureFusion(nn.Module):
+    def __init__(self, inchannels, outchannels):
+        super(FeatureFusion, self).__init__()
+        self.conv = ResidualConv(inchannels=inchannels)
+        # NN.BatchNorm2d
+        self.up = nn.Sequential(ResidualConv(inchannels=inchannels),
+                                nn.ConvTranspose2d(in_channels=inchannels, out_channels=outchannels, kernel_size=3,
+                                                   stride=2, padding=1, output_padding=1),
+                                nn.BatchNorm2d(num_features=outchannels),
+                                nn.ReLU(inplace=True))
+
+    def forward(self, lowfeat, highfeat):
+        return self.up(highfeat + self.conv(lowfeat))
+
+    def init_params(self):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                init.normal_(m.weight, std=0.01)
+                # init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.ConvTranspose2d):
+                # init.kaiming_normal_(m.weight, mode='fan_out')
+                init.normal_(m.weight, std=0.01)
+                # init.xavier_normal_(m.weight)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+            elif isinstance(m, nn.BatchNorm2d):  # NN.BatchNorm2d
+                init.constant_(m.weight, 1)
+                init.constant_(m.bias, 0)
+            elif isinstance(m, nn.Linear):
+                init.normal_(m.weight, std=0.01)
+                if m.bias is not None:
+                    init.constant_(m.bias, 0)
+
+
+class SenceUnderstand(nn.Module):
+    def __init__(self, channels):
+        super(SenceUnderstand, self).__init__()
+        self.channels = channels
+        self.conv1 = nn.Sequential(nn.Conv2d(in_channels=512, out_channels=512, kernel_size=3, padding=1),
+                                   nn.ReLU(inplace=True))
+        self.pool = nn.AdaptiveAvgPool2d(8)
+        self.fc = nn.Sequential(nn.Linear(512 * 8 * 8, self.channels),
+                                nn.ReLU(inplace=True))
+        self.conv2 = nn.Sequential(
+            nn.Conv2d(in_channels=self.channels, out_channels=self.channels, kernel_size=1, padding=0),
+            nn.ReLU(inplace=True))
+        self.initial_params()
+
+    def forward(self, x):
+        n, c, h, w = x.size()
+        x = self.conv1(x)
+        x = self.pool(x)
+        x = x.view(n, -1)
+        x = self.fc(x)
+        x = x.view(n, self.channels, 1, 1)
+        x = self.conv2(x)
+        x = x.repeat(1, 1, h, w)
+        return x
+
+    def initial_params(self, dev=0.01):
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                # print torch.sum(m.weight)
+                m.weight.data.normal_(0, dev)
+                if m.bias is not None:
+                    m.bias.data.fill_(0)
+            elif isinstance(m, nn.ConvTranspose2d):
+                # print torch.sum(m.weight)
+                m.weight.data.normal_(0, dev)
+                if m.bias is not None:
+                    m.bias.data.fill_(0)
+            elif isinstance(m, nn.Linear):
+                m.weight.data.normal_(0, dev)
+
+
+if __name__ == '__main__':
+    net = DepthNet(depth=50, pretrained=True)
+    print(net)
+    inputs = torch.ones(4,3,128,128)
+    out = net(inputs)
+    print(out.size())
+

negi/repos/controlnet_aux/src/controlnet_aux/leres/pix2pix/LICENSE

@@ -0,0 +1,19 @@
+https://github.com/compphoto/BoostingMonocularDepth
+
+Copyright 2021, Seyed Mahdi Hosseini Miangoleh, Sebastian Dille, Computational Photography Laboratory. All rights reserved.
+
+This software is for academic use only. A redistribution of this 
+software, with or  without modifications, has to be for academic 
+use only, while giving the appropriate credit to the original 
+authors of the software. The methods implemented as a part of 
+this software may be covered under patents or patent applications.
+
+THIS SOFTWARE IS PROVIDED BY THE AUTHOR ''AS IS'' AND ANY EXPRESS OR IMPLIED
+WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR
+CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

negi/repos/controlnet_aux/src/controlnet_aux/leres/pix2pix/options/__init__.py

@@ -0,0 +1 @@
+"""This package options includes option modules: training options, test options, and basic options (used in both training and test)."""

negi/repos/controlnet_aux/src/controlnet_aux/leres/pix2pix/options/base_options.py

@@ -0,0 +1,156 @@
+import argparse
+import os
+from ...pix2pix.util import util
+# import torch
+from ...pix2pix import models
+# import pix2pix.data
+import numpy as np
+
+class BaseOptions():
+    """This class defines options used during both training and test time.
+
+    It also implements several helper functions such as parsing, printing, and saving the options.
+    It also gathers additional options defined in <modify_commandline_options> functions in both dataset class and model class.
+    """
+
+    def __init__(self):
+        """Reset the class; indicates the class hasn't been initailized"""
+        self.initialized = False
+
+    def initialize(self, parser):
+        """Define the common options that are used in both training and test."""
+        # basic parameters
+        parser.add_argument('--dataroot', help='path to images (should have subfolders trainA, trainB, valA, valB, etc)')
+        parser.add_argument('--name', type=str, default='void', help='mahdi_unet_new, scaled_unet')
+        parser.add_argument('--gpu_ids', type=str, default='0', help='gpu ids: e.g. 0  0,1,2, 0,2. use -1 for CPU')
+        parser.add_argument('--checkpoints_dir', type=str, default='./pix2pix/checkpoints', help='models are saved here')
+        # model parameters
+        parser.add_argument('--model', type=str, default='cycle_gan', help='chooses which model to use. [cycle_gan | pix2pix | test | colorization]')
+        parser.add_argument('--input_nc', type=int, default=2, help='# of input image channels: 3 for RGB and 1 for grayscale')
+        parser.add_argument('--output_nc', type=int, default=1, help='# of output image channels: 3 for RGB and 1 for grayscale')
+        parser.add_argument('--ngf', type=int, default=64, help='# of gen filters in the last conv layer')
+        parser.add_argument('--ndf', type=int, default=64, help='# of discrim filters in the first conv layer')
+        parser.add_argument('--netD', type=str, default='basic', help='specify discriminator architecture [basic | n_layers | pixel]. The basic model is a 70x70 PatchGAN. n_layers allows you to specify the layers in the discriminator')
+        parser.add_argument('--netG', type=str, default='resnet_9blocks', help='specify generator architecture [resnet_9blocks | resnet_6blocks | unet_256 | unet_128]')
+        parser.add_argument('--n_layers_D', type=int, default=3, help='only used if netD==n_layers')
+        parser.add_argument('--norm', type=str, default='instance', help='instance normalization or batch normalization [instance | batch | none]')
+        parser.add_argument('--init_type', type=str, default='normal', help='network initialization [normal | xavier | kaiming | orthogonal]')
+        parser.add_argument('--init_gain', type=float, default=0.02, help='scaling factor for normal, xavier and orthogonal.')
+        parser.add_argument('--no_dropout', action='store_true', help='no dropout for the generator')
+        # dataset parameters
+        parser.add_argument('--dataset_mode', type=str, default='unaligned', help='chooses how datasets are loaded. [unaligned | aligned | single | colorization]')
+        parser.add_argument('--direction', type=str, default='AtoB', help='AtoB or BtoA')
+        parser.add_argument('--serial_batches', action='store_true', help='if true, takes images in order to make batches, otherwise takes them randomly')
+        parser.add_argument('--num_threads', default=4, type=int, help='# threads for loading data')
+        parser.add_argument('--batch_size', type=int, default=1, help='input batch size')
+        parser.add_argument('--load_size', type=int, default=672, help='scale images to this size')
+        parser.add_argument('--crop_size', type=int, default=672, help='then crop to this size')
+        parser.add_argument('--max_dataset_size', type=int, default=10000, help='Maximum number of samples allowed per dataset. If the dataset directory contains more than max_dataset_size, only a subset is loaded.')
+        parser.add_argument('--preprocess', type=str, default='resize_and_crop', help='scaling and cropping of images at load time [resize_and_crop | crop | scale_width | scale_width_and_crop | none]')
+        parser.add_argument('--no_flip', action='store_true', help='if specified, do not flip the images for data augmentation')
+        parser.add_argument('--display_winsize', type=int, default=256, help='display window size for both visdom and HTML')
+        # additional parameters
+        parser.add_argument('--epoch', type=str, default='latest', help='which epoch to load? set to latest to use latest cached model')
+        parser.add_argument('--load_iter', type=int, default='0', help='which iteration to load? if load_iter > 0, the code will load models by iter_[load_iter]; otherwise, the code will load models by [epoch]')
+        parser.add_argument('--verbose', action='store_true', help='if specified, print more debugging information')
+        parser.add_argument('--suffix', default='', type=str, help='customized suffix: opt.name = opt.name + suffix: e.g., {model}_{netG}_size{load_size}')
+
+        parser.add_argument('--data_dir', type=str, required=False,
+                            help='input files directory images can be .png .jpg .tiff')
+        parser.add_argument('--output_dir', type=str, required=False,
+                            help='result dir. result depth will be png. vides are JMPG as avi')
+        parser.add_argument('--savecrops', type=int, required=False)
+        parser.add_argument('--savewholeest', type=int, required=False)
+        parser.add_argument('--output_resolution', type=int, required=False,
+                            help='0 for no restriction 1 for resize to input size')
+        parser.add_argument('--net_receptive_field_size', type=int, required=False)
+        parser.add_argument('--pix2pixsize', type=int, required=False)
+        parser.add_argument('--generatevideo', type=int, required=False)
+        parser.add_argument('--depthNet', type=int, required=False, help='0: midas 1:strurturedRL')
+        parser.add_argument('--R0', action='store_true')
+        parser.add_argument('--R20', action='store_true')
+        parser.add_argument('--Final', action='store_true')
+        parser.add_argument('--colorize_results', action='store_true')
+        parser.add_argument('--max_res', type=float, default=np.inf)
+
+        self.initialized = True
+        return parser
+
+    def gather_options(self):
+        """Initialize our parser with basic options(only once).
+        Add additional model-specific and dataset-specific options.
+        These options are defined in the <modify_commandline_options> function
+        in model and dataset classes.
+        """
+        if not self.initialized:  # check if it has been initialized
+            parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
+            parser = self.initialize(parser)
+
+        # get the basic options
+        opt, _ = parser.parse_known_args()
+
+        # modify model-related parser options
+        model_name = opt.model
+        model_option_setter = models.get_option_setter(model_name)
+        parser = model_option_setter(parser, self.isTrain)
+        opt, _ = parser.parse_known_args()  # parse again with new defaults
+
+        # modify dataset-related parser options
+        # dataset_name = opt.dataset_mode
+        # dataset_option_setter = pix2pix.data.get_option_setter(dataset_name)
+        # parser = dataset_option_setter(parser, self.isTrain)
+
+        # save and return the parser
+        self.parser = parser
+        #return parser.parse_args() #EVIL
+        return opt
+
+    def print_options(self, opt):
+        """Print and save options
+
+        It will print both current options and default values(if different).
+        It will save options into a text file / [checkpoints_dir] / opt.txt
+        """
+        message = ''
+        message += '----------------- Options ---------------\n'
+        for k, v in sorted(vars(opt).items()):
+            comment = ''
+            default = self.parser.get_default(k)
+            if v != default:
+                comment = '\t[default: %s]' % str(default)
+            message += '{:>25}: {:<30}{}\n'.format(str(k), str(v), comment)
+        message += '----------------- End -------------------'
+        print(message)
+
+        # save to the disk
+        expr_dir = os.path.join(opt.checkpoints_dir, opt.name)
+        util.mkdirs(expr_dir)
+        file_name = os.path.join(expr_dir, '{}_opt.txt'.format(opt.phase))
+        with open(file_name, 'wt') as opt_file:
+            opt_file.write(message)
+            opt_file.write('\n')
+
+    def parse(self):
+        """Parse our options, create checkpoints directory suffix, and set up gpu device."""
+        opt = self.gather_options()
+        opt.isTrain = self.isTrain   # train or test
+
+        # process opt.suffix
+        if opt.suffix:
+            suffix = ('_' + opt.suffix.format(**vars(opt))) if opt.suffix != '' else ''
+            opt.name = opt.name + suffix
+
+        #self.print_options(opt)
+
+        # set gpu ids
+        str_ids = opt.gpu_ids.split(',')
+        opt.gpu_ids = []
+        for str_id in str_ids:
+            id = int(str_id)
+            if id >= 0:
+                opt.gpu_ids.append(id)
+        #if len(opt.gpu_ids) > 0:
+        #    torch.cuda.set_device(opt.gpu_ids[0])
+
+        self.opt = opt
+        return self.opt

negi/repos/controlnet_aux/src/controlnet_aux/leres/pix2pix/options/test_options.py

@@ -0,0 +1,22 @@
+from .base_options import BaseOptions
+
+
+class TestOptions(BaseOptions):
+    """This class includes test options.
+
+    It also includes shared options defined in BaseOptions.
+    """
+
+    def initialize(self, parser):
+        parser = BaseOptions.initialize(self, parser)  # define shared options
+        parser.add_argument('--aspect_ratio', type=float, default=1.0, help='aspect ratio of result images')
+        parser.add_argument('--phase', type=str, default='test', help='train, val, test, etc')
+        # Dropout and Batchnorm has different behavioir during training and test.
+        parser.add_argument('--eval', action='store_true', help='use eval mode during test time.')
+        parser.add_argument('--num_test', type=int, default=50, help='how many test images to run')
+        # rewrite devalue values
+        parser.set_defaults(model='pix2pix4depth')
+        # To avoid cropping, the load_size should be the same as crop_size
+        parser.set_defaults(load_size=parser.get_default('crop_size'))
+        self.isTrain = False
+        return parser

negi/repos/controlnet_aux/src/controlnet_aux/leres/pix2pix/util/__init__.py

@@ -0,0 +1 @@
+"""This package includes a miscellaneous collection of useful helper functions."""

negi/repos/controlnet_aux/src/controlnet_aux/leres/pix2pix/util/util.py

@@ -0,0 +1,105 @@
+"""This module contains simple helper functions """
+from __future__ import print_function
+import torch
+import numpy as np
+from PIL import Image
+import os
+
+
+def tensor2im(input_image, imtype=np.uint16):
+    """"Converts a Tensor array into a numpy image array.
+
+    Parameters:
+        input_image (tensor) --  the input image tensor array
+        imtype (type)        --  the desired type of the converted numpy array
+    """
+    if not isinstance(input_image, np.ndarray):
+        if isinstance(input_image, torch.Tensor):  # get the data from a variable
+            image_tensor = input_image.data
+        else:
+            return input_image
+        image_numpy = torch.squeeze(image_tensor).cpu().numpy()  # convert it into a numpy array
+        image_numpy = (image_numpy + 1) / 2.0 * (2**16-1) #
+    else:  # if it is a numpy array, do nothing
+        image_numpy = input_image
+    return image_numpy.astype(imtype)
+
+
+def diagnose_network(net, name='network'):
+    """Calculate and print the mean of average absolute(gradients)
+
+    Parameters:
+        net (torch network) -- Torch network
+        name (str) -- the name of the network
+    """
+    mean = 0.0
+    count = 0
+    for param in net.parameters():
+        if param.grad is not None:
+            mean += torch.mean(torch.abs(param.grad.data))
+            count += 1
+    if count > 0:
+        mean = mean / count
+    print(name)
+    print(mean)
+
+
+def save_image(image_numpy, image_path, aspect_ratio=1.0):
+    """Save a numpy image to the disk
+
+    Parameters:
+        image_numpy (numpy array) -- input numpy array
+        image_path (str)          -- the path of the image
+    """
+    image_pil = Image.fromarray(image_numpy)
+
+    image_pil = image_pil.convert('I;16')
+
+    # image_pil = Image.fromarray(image_numpy)
+    # h, w, _ = image_numpy.shape
+    #
+    # if aspect_ratio > 1.0:
+    #     image_pil = image_pil.resize((h, int(w * aspect_ratio)), Image.BICUBIC)
+    # if aspect_ratio < 1.0:
+    #     image_pil = image_pil.resize((int(h / aspect_ratio), w), Image.BICUBIC)
+
+    image_pil.save(image_path)
+
+
+def print_numpy(x, val=True, shp=False):
+    """Print the mean, min, max, median, std, and size of a numpy array
+
+    Parameters:
+        val (bool) -- if print the values of the numpy array
+        shp (bool) -- if print the shape of the numpy array
+    """
+    x = x.astype(np.float64)
+    if shp:
+        print('shape,', x.shape)
+    if val:
+        x = x.flatten()
+        print('mean = %3.3f, min = %3.3f, max = %3.3f, median = %3.3f, std=%3.3f' % (
+            np.mean(x), np.min(x), np.max(x), np.median(x), np.std(x)))
+
+
+def mkdirs(paths):
+    """create empty directories if they don't exist
+
+    Parameters:
+        paths (str list) -- a list of directory paths
+    """
+    if isinstance(paths, list) and not isinstance(paths, str):
+        for path in paths:
+            mkdir(path)
+    else:
+        mkdir(paths)
+
+
+def mkdir(path):
+    """create a single empty directory if it didn't exist
+
+    Parameters:
+        path (str) -- a single directory path
+    """
+    if not os.path.exists(path):
+        os.makedirs(path)

negi/repos/controlnet_aux/src/controlnet_aux/lineart/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2022 Caroline Chan
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

negi/repos/controlnet_aux/src/controlnet_aux/lineart/__init__.py

@@ -0,0 +1,167 @@
+import os
+import warnings
+
+import cv2
+import numpy as np
+import torch
+import torch.nn as nn
+from einops import rearrange
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from ..util import HWC3, resize_image
+
+norm_layer = nn.InstanceNorm2d
+
+
+class ResidualBlock(nn.Module):
+    def __init__(self, in_features):
+        super(ResidualBlock, self).__init__()
+
+        conv_block = [  nn.ReflectionPad2d(1),
+                        nn.Conv2d(in_features, in_features, 3),
+                        norm_layer(in_features),
+                        nn.ReLU(inplace=True),
+                        nn.ReflectionPad2d(1),
+                        nn.Conv2d(in_features, in_features, 3),
+                        norm_layer(in_features)
+                        ]
+
+        self.conv_block = nn.Sequential(*conv_block)
+
+    def forward(self, x):
+        return x + self.conv_block(x)
+
+
+class Generator(nn.Module):
+    def __init__(self, input_nc, output_nc, n_residual_blocks=9, sigmoid=True):
+        super(Generator, self).__init__()
+
+        # Initial convolution block
+        model0 = [   nn.ReflectionPad2d(3),
+                    nn.Conv2d(input_nc, 64, 7),
+                    norm_layer(64),
+                    nn.ReLU(inplace=True) ]
+        self.model0 = nn.Sequential(*model0)
+
+        # Downsampling
+        model1 = []
+        in_features = 64
+        out_features = in_features*2
+        for _ in range(2):
+            model1 += [  nn.Conv2d(in_features, out_features, 3, stride=2, padding=1),
+                        norm_layer(out_features),
+                        nn.ReLU(inplace=True) ]
+            in_features = out_features
+            out_features = in_features*2
+        self.model1 = nn.Sequential(*model1)
+
+        model2 = []
+        # Residual blocks
+        for _ in range(n_residual_blocks):
+            model2 += [ResidualBlock(in_features)]
+        self.model2 = nn.Sequential(*model2)
+
+        # Upsampling
+        model3 = []
+        out_features = in_features//2
+        for _ in range(2):
+            model3 += [  nn.ConvTranspose2d(in_features, out_features, 3, stride=2, padding=1, output_padding=1),
+                        norm_layer(out_features),
+                        nn.ReLU(inplace=True) ]
+            in_features = out_features
+            out_features = in_features//2
+        self.model3 = nn.Sequential(*model3)
+
+        # Output layer
+        model4 = [  nn.ReflectionPad2d(3),
+                        nn.Conv2d(64, output_nc, 7)]
+        if sigmoid:
+            model4 += [nn.Sigmoid()]
+
+        self.model4 = nn.Sequential(*model4)
+
+    def forward(self, x, cond=None):
+        out = self.model0(x)
+        out = self.model1(out)
+        out = self.model2(out)
+        out = self.model3(out)
+        out = self.model4(out)
+
+        return out
+
+
+class LineartDetector:
+    def __init__(self, model, coarse_model):
+        self.model = model
+        self.model_coarse = coarse_model
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_or_path, filename=None, coarse_filename=None, cache_dir=None):
+        filename = filename or "sk_model.pth"
+        coarse_filename = coarse_filename or "sk_model2.pth"
+
+        if os.path.isdir(pretrained_model_or_path):
+            model_path = os.path.join(pretrained_model_or_path, filename)
+            coarse_model_path = os.path.join(pretrained_model_or_path, coarse_filename)
+        else:
+            model_path = hf_hub_download(pretrained_model_or_path, filename, cache_dir=cache_dir)
+            coarse_model_path = hf_hub_download(pretrained_model_or_path, coarse_filename, cache_dir=cache_dir)
+
+        model = Generator(3, 1, 3)
+        model.load_state_dict(torch.load(model_path, map_location=torch.device('cpu')))
+        model.eval()
+
+        coarse_model = Generator(3, 1, 3)
+        coarse_model.load_state_dict(torch.load(coarse_model_path, map_location=torch.device('cpu')))
+        coarse_model.eval()
+
+        return cls(model, coarse_model)
+    
+    def to(self, device):
+        self.model.to(device)
+        self.model_coarse.to(device)
+        return self
+    
+    def __call__(self, input_image, coarse=False, detect_resolution=512, image_resolution=512, output_type="pil", **kwargs):
+        if "return_pil" in kwargs:
+            warnings.warn("return_pil is deprecated. Use output_type instead.", DeprecationWarning)
+            output_type = "pil" if kwargs["return_pil"] else "np"
+        if type(output_type) is bool:
+            warnings.warn("Passing `True` or `False` to `output_type` is deprecated and will raise an error in future versions")
+            if output_type:
+                output_type = "pil"
+
+        device = next(iter(self.model.parameters())).device
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+
+        model = self.model_coarse if coarse else self.model
+        assert input_image.ndim == 3
+        image = input_image
+        with torch.no_grad():
+            image = torch.from_numpy(image).float().to(device)
+            image = image / 255.0
+            image = rearrange(image, 'h w c -> 1 c h w')
+            line = model(image)[0][0]
+
+            line = line.cpu().numpy()
+            line = (line * 255.0).clip(0, 255).astype(np.uint8)
+
+        detected_map = line
+
+        detected_map = HWC3(detected_map)
+
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+        detected_map = 255 - detected_map
+        
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+
+        return detected_map

negi/repos/controlnet_aux/src/controlnet_aux/lineart_anime/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2022 Caroline Chan
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

negi/repos/controlnet_aux/src/controlnet_aux/lineart_anime/__init__.py

@@ -0,0 +1,189 @@
+import functools
+import os
+import warnings
+
+import cv2
+import numpy as np
+import torch
+import torch.nn as nn
+from einops import rearrange
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from ..util import HWC3, resize_image
+
+
+class UnetGenerator(nn.Module):
+    """Create a Unet-based generator"""
+
+    def __init__(self, input_nc, output_nc, num_downs, ngf=64, norm_layer=nn.BatchNorm2d, use_dropout=False):
+        """Construct a Unet generator
+        Parameters:
+            input_nc (int)  -- the number of channels in input images
+            output_nc (int) -- the number of channels in output images
+            num_downs (int) -- the number of downsamplings in UNet. For example, # if |num_downs| == 7,
+                                image of size 128x128 will become of size 1x1 # at the bottleneck
+            ngf (int)       -- the number of filters in the last conv layer
+            norm_layer      -- normalization layer
+        We construct the U-Net from the innermost layer to the outermost layer.
+        It is a recursive process.
+        """
+        super(UnetGenerator, self).__init__()
+        # construct unet structure
+        unet_block = UnetSkipConnectionBlock(ngf * 8, ngf * 8, input_nc=None, submodule=None, norm_layer=norm_layer, innermost=True)  # add the innermost layer
+        for _ in range(num_downs - 5):          # add intermediate layers with ngf * 8 filters
+            unet_block = UnetSkipConnectionBlock(ngf * 8, ngf * 8, input_nc=None, submodule=unet_block, norm_layer=norm_layer, use_dropout=use_dropout)
+        # gradually reduce the number of filters from ngf * 8 to ngf
+        unet_block = UnetSkipConnectionBlock(ngf * 4, ngf * 8, input_nc=None, submodule=unet_block, norm_layer=norm_layer)
+        unet_block = UnetSkipConnectionBlock(ngf * 2, ngf * 4, input_nc=None, submodule=unet_block, norm_layer=norm_layer)
+        unet_block = UnetSkipConnectionBlock(ngf, ngf * 2, input_nc=None, submodule=unet_block, norm_layer=norm_layer)
+        self.model = UnetSkipConnectionBlock(output_nc, ngf, input_nc=input_nc, submodule=unet_block, outermost=True, norm_layer=norm_layer)  # add the outermost layer
+
+    def forward(self, input):
+        """Standard forward"""
+        return self.model(input)
+
+
+class UnetSkipConnectionBlock(nn.Module):
+    """Defines the Unet submodule with skip connection.
+        X -------------------identity----------------------
+        |-- downsampling -- |submodule| -- upsampling --|
+    """
+
+    def __init__(self, outer_nc, inner_nc, input_nc=None,
+                 submodule=None, outermost=False, innermost=False, norm_layer=nn.BatchNorm2d, use_dropout=False):
+        """Construct a Unet submodule with skip connections.
+        Parameters:
+            outer_nc (int) -- the number of filters in the outer conv layer
+            inner_nc (int) -- the number of filters in the inner conv layer
+            input_nc (int) -- the number of channels in input images/features
+            submodule (UnetSkipConnectionBlock) -- previously defined submodules
+            outermost (bool)    -- if this module is the outermost module
+            innermost (bool)    -- if this module is the innermost module
+            norm_layer          -- normalization layer
+            use_dropout (bool)  -- if use dropout layers.
+        """
+        super(UnetSkipConnectionBlock, self).__init__()
+        self.outermost = outermost
+        if type(norm_layer) == functools.partial:
+            use_bias = norm_layer.func == nn.InstanceNorm2d
+        else:
+            use_bias = norm_layer == nn.InstanceNorm2d
+        if input_nc is None:
+            input_nc = outer_nc
+        downconv = nn.Conv2d(input_nc, inner_nc, kernel_size=4,
+                             stride=2, padding=1, bias=use_bias)
+        downrelu = nn.LeakyReLU(0.2, True)
+        downnorm = norm_layer(inner_nc)
+        uprelu = nn.ReLU(True)
+        upnorm = norm_layer(outer_nc)
+
+        if outermost:
+            upconv = nn.ConvTranspose2d(inner_nc * 2, outer_nc,
+                                        kernel_size=4, stride=2,
+                                        padding=1)
+            down = [downconv]
+            up = [uprelu, upconv, nn.Tanh()]
+            model = down + [submodule] + up
+        elif innermost:
+            upconv = nn.ConvTranspose2d(inner_nc, outer_nc,
+                                        kernel_size=4, stride=2,
+                                        padding=1, bias=use_bias)
+            down = [downrelu, downconv]
+            up = [uprelu, upconv, upnorm]
+            model = down + up
+        else:
+            upconv = nn.ConvTranspose2d(inner_nc * 2, outer_nc,
+                                        kernel_size=4, stride=2,
+                                        padding=1, bias=use_bias)
+            down = [downrelu, downconv, downnorm]
+            up = [uprelu, upconv, upnorm]
+
+            if use_dropout:
+                model = down + [submodule] + up + [nn.Dropout(0.5)]
+            else:
+                model = down + [submodule] + up
+
+        self.model = nn.Sequential(*model)
+
+    def forward(self, x):
+        if self.outermost:
+            return self.model(x)
+        else:   # add skip connections
+            return torch.cat([x, self.model(x)], 1)
+
+
+class LineartAnimeDetector:
+    def __init__(self, model):
+        self.model = model
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_or_path, filename=None, cache_dir=None):
+        filename = filename or "netG.pth"
+
+        if os.path.isdir(pretrained_model_or_path):
+            model_path = os.path.join(pretrained_model_or_path, filename)
+        else:
+            model_path = hf_hub_download(pretrained_model_or_path, filename, cache_dir=cache_dir)
+
+        norm_layer = functools.partial(nn.InstanceNorm2d, affine=False, track_running_stats=False)
+        net = UnetGenerator(3, 1, 8, 64, norm_layer=norm_layer, use_dropout=False)
+        ckpt = torch.load(model_path)
+        for key in list(ckpt.keys()):
+            if 'module.' in key:
+                ckpt[key.replace('module.', '')] = ckpt[key]
+                del ckpt[key]
+        net.load_state_dict(ckpt)
+        net.eval()
+
+        return cls(net)
+
+    def to(self, device):
+        self.model.to(device)
+        return self
+    
+    def __call__(self, input_image, detect_resolution=512, image_resolution=512, output_type="pil", **kwargs):
+        if "return_pil" in kwargs:
+            warnings.warn("return_pil is deprecated. Use output_type instead.", DeprecationWarning)
+            output_type = "pil" if kwargs["return_pil"] else "np"
+        if type(output_type) is bool:
+            warnings.warn("Passing `True` or `False` to `output_type` is deprecated and will raise an error in future versions")
+            if output_type:
+                output_type = "pil"
+
+        device = next(iter(self.model.parameters())).device
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+
+        H, W, C = input_image.shape
+        Hn = 256 * int(np.ceil(float(H) / 256.0))
+        Wn = 256 * int(np.ceil(float(W) / 256.0))
+        img = cv2.resize(input_image, (Wn, Hn), interpolation=cv2.INTER_CUBIC)
+        with torch.no_grad():
+            image_feed = torch.from_numpy(img).float().to(device)
+            image_feed = image_feed / 127.5 - 1.0
+            image_feed = rearrange(image_feed, 'h w c -> 1 c h w')
+
+            line = self.model(image_feed)[0, 0] * 127.5 + 127.5
+            line = line.cpu().numpy()
+
+            line = cv2.resize(line, (W, H), interpolation=cv2.INTER_CUBIC)
+            line = line.clip(0, 255).astype(np.uint8)
+
+        detected_map = line
+
+        detected_map = HWC3(detected_map)
+
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+        detected_map = 255 - detected_map
+        
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+
+        return detected_map

negi/repos/controlnet_aux/src/controlnet_aux/mediapipe_face/__init__.py

@@ -0,0 +1,53 @@
+import warnings
+from typing import Union
+
+import cv2
+import numpy as np
+from PIL import Image
+
+from ..util import HWC3, resize_image
+from .mediapipe_face_common import generate_annotation
+
+
+class MediapipeFaceDetector:
+    def __call__(self,
+                 input_image: Union[np.ndarray, Image.Image] = None,
+                 max_faces: int = 1,
+                 min_confidence: float = 0.5,
+                 output_type: str = "pil",
+                 detect_resolution: int = 512,
+                 image_resolution: int = 512,
+                 **kwargs):
+
+        if "image" in kwargs:
+            warnings.warn("image is deprecated, please use `input_image=...` instead.", DeprecationWarning)
+            input_image = kwargs.pop("image")        
+        if input_image is None:
+            raise ValueError("input_image must be defined.")
+
+        if "return_pil" in kwargs:
+            warnings.warn("return_pil is deprecated. Use output_type instead.", DeprecationWarning)
+            output_type = "pil" if kwargs["return_pil"] else "np"
+        if type(output_type) is bool:
+            warnings.warn("Passing `True` or `False` to `output_type` is deprecated and will raise an error in future versions")
+            if output_type:
+                output_type = "pil"
+
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+
+        detected_map = generate_annotation(input_image, max_faces, min_confidence)
+        detected_map = HWC3(detected_map)
+
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+        
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+            
+        return detected_map

negi/repos/controlnet_aux/src/controlnet_aux/mediapipe_face/mediapipe_face_common.py

@@ -0,0 +1,164 @@
+from typing import Mapping
+import warnings
+
+try:
+    import mediapipe as mp
+except ImportError:
+    warnings.warn(
+        "The module 'mediapipe' is not installed. The package will have limited functionality. Please install it using the command: pip install 'mediapipe'"
+    )
+
+    mp = None
+
+import numpy
+
+if mp:
+    mp_drawing = mp.solutions.drawing_utils
+    mp_drawing_styles = mp.solutions.drawing_styles
+    mp_face_detection = mp.solutions.face_detection  # Only for counting faces.
+    mp_face_mesh = mp.solutions.face_mesh
+    mp_face_connections = mp.solutions.face_mesh_connections.FACEMESH_TESSELATION
+    mp_hand_connections = mp.solutions.hands_connections.HAND_CONNECTIONS
+    mp_body_connections = mp.solutions.pose_connections.POSE_CONNECTIONS
+
+    DrawingSpec = mp.solutions.drawing_styles.DrawingSpec
+    PoseLandmark = mp.solutions.drawing_styles.PoseLandmark
+
+    min_face_size_pixels: int = 64
+    f_thick = 2
+    f_rad = 1
+    right_iris_draw = DrawingSpec(color=(10, 200, 250), thickness=f_thick, circle_radius=f_rad)
+    right_eye_draw = DrawingSpec(color=(10, 200, 180), thickness=f_thick, circle_radius=f_rad)
+    right_eyebrow_draw = DrawingSpec(color=(10, 220, 180), thickness=f_thick, circle_radius=f_rad)
+    left_iris_draw = DrawingSpec(color=(250, 200, 10), thickness=f_thick, circle_radius=f_rad)
+    left_eye_draw = DrawingSpec(color=(180, 200, 10), thickness=f_thick, circle_radius=f_rad)
+    left_eyebrow_draw = DrawingSpec(color=(180, 220, 10), thickness=f_thick, circle_radius=f_rad)
+    mouth_draw = DrawingSpec(color=(10, 180, 10), thickness=f_thick, circle_radius=f_rad)
+    head_draw = DrawingSpec(color=(10, 200, 10), thickness=f_thick, circle_radius=f_rad)
+
+    # mp_face_mesh.FACEMESH_CONTOURS has all the items we care about.
+    face_connection_spec = {}
+    for edge in mp_face_mesh.FACEMESH_FACE_OVAL:
+        face_connection_spec[edge] = head_draw
+    for edge in mp_face_mesh.FACEMESH_LEFT_EYE:
+        face_connection_spec[edge] = left_eye_draw
+    for edge in mp_face_mesh.FACEMESH_LEFT_EYEBROW:
+        face_connection_spec[edge] = left_eyebrow_draw
+    # for edge in mp_face_mesh.FACEMESH_LEFT_IRIS:
+    #    face_connection_spec[edge] = left_iris_draw
+    for edge in mp_face_mesh.FACEMESH_RIGHT_EYE:
+        face_connection_spec[edge] = right_eye_draw
+    for edge in mp_face_mesh.FACEMESH_RIGHT_EYEBROW:
+        face_connection_spec[edge] = right_eyebrow_draw
+    # for edge in mp_face_mesh.FACEMESH_RIGHT_IRIS:
+    #    face_connection_spec[edge] = right_iris_draw
+    for edge in mp_face_mesh.FACEMESH_LIPS:
+        face_connection_spec[edge] = mouth_draw
+    iris_landmark_spec = {468: right_iris_draw, 473: left_iris_draw}
+
+
+def draw_pupils(image, landmark_list, drawing_spec, halfwidth: int = 2):
+    """We have a custom function to draw the pupils because the mp.draw_landmarks method requires a parameter for all
+    landmarks.  Until our PR is merged into mediapipe, we need this separate method."""
+    if len(image.shape) != 3:
+        raise ValueError("Input image must be H,W,C.")
+    image_rows, image_cols, image_channels = image.shape
+    if image_channels != 3:  # BGR channels
+        raise ValueError('Input image must contain three channel bgr data.')
+    for idx, landmark in enumerate(landmark_list.landmark):
+        if (
+                (landmark.HasField('visibility') and landmark.visibility < 0.9) or
+                (landmark.HasField('presence') and landmark.presence < 0.5)
+        ):
+            continue
+        if landmark.x >= 1.0 or landmark.x < 0 or landmark.y >= 1.0 or landmark.y < 0:
+            continue
+        image_x = int(image_cols*landmark.x)
+        image_y = int(image_rows*landmark.y)
+        draw_color = None
+        if isinstance(drawing_spec, Mapping):
+            if drawing_spec.get(idx) is None:
+                continue
+            else:
+                draw_color = drawing_spec[idx].color
+        elif isinstance(drawing_spec, DrawingSpec):
+            draw_color = drawing_spec.color
+        image[image_y-halfwidth:image_y+halfwidth, image_x-halfwidth:image_x+halfwidth, :] = draw_color
+
+
+def reverse_channels(image):
+    """Given a numpy array in RGB form, convert to BGR.  Will also convert from BGR to RGB."""
+    # im[:,:,::-1] is a neat hack to convert BGR to RGB by reversing the indexing order.
+    # im[:,:,::[2,1,0]] would also work but makes a copy of the data.
+    return image[:, :, ::-1]
+
+
+def generate_annotation(
+        img_rgb,
+        max_faces: int,
+        min_confidence: float
+):
+    """
+    Find up to 'max_faces' inside the provided input image.
+    If min_face_size_pixels is provided and nonzero it will be used to filter faces that occupy less than this many
+    pixels in the image.
+    """
+    with mp_face_mesh.FaceMesh(
+            static_image_mode=True,
+            max_num_faces=max_faces,
+            refine_landmarks=True,
+            min_detection_confidence=min_confidence,
+    ) as facemesh:
+        img_height, img_width, img_channels = img_rgb.shape
+        assert(img_channels == 3)
+
+        results = facemesh.process(img_rgb).multi_face_landmarks
+
+        if results is None:
+            print("No faces detected in controlnet image for Mediapipe face annotator.")
+            return numpy.zeros_like(img_rgb)
+
+        # Filter faces that are too small
+        filtered_landmarks = []
+        for lm in results:
+            landmarks = lm.landmark
+            face_rect = [
+                landmarks[0].x,
+                landmarks[0].y,
+                landmarks[0].x,
+                landmarks[0].y,
+            ]  # Left, up, right, down.
+            for i in range(len(landmarks)):
+                face_rect[0] = min(face_rect[0], landmarks[i].x)
+                face_rect[1] = min(face_rect[1], landmarks[i].y)
+                face_rect[2] = max(face_rect[2], landmarks[i].x)
+                face_rect[3] = max(face_rect[3], landmarks[i].y)
+            if min_face_size_pixels > 0:
+                face_width = abs(face_rect[2] - face_rect[0])
+                face_height = abs(face_rect[3] - face_rect[1])
+                face_width_pixels = face_width * img_width
+                face_height_pixels = face_height * img_height
+                face_size = min(face_width_pixels, face_height_pixels)
+                if face_size >= min_face_size_pixels:
+                    filtered_landmarks.append(lm)
+            else:
+                filtered_landmarks.append(lm)
+
+        # Annotations are drawn in BGR for some reason, but we don't need to flip a zero-filled image at the start.
+        empty = numpy.zeros_like(img_rgb)
+
+        # Draw detected faces:
+        for face_landmarks in filtered_landmarks:
+            mp_drawing.draw_landmarks(
+                empty,
+                face_landmarks,
+                connections=face_connection_spec.keys(),
+                landmark_drawing_spec=None,
+                connection_drawing_spec=face_connection_spec
+            )
+            draw_pupils(empty, face_landmarks, iris_landmark_spec, 2)
+
+        # Flip BGR back to RGB.
+        empty = reverse_channels(empty).copy()
+
+        return empty

negi/repos/controlnet_aux/src/controlnet_aux/midas/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2019 Intel ISL (Intel Intelligent Systems Lab)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

negi/repos/controlnet_aux/src/controlnet_aux/midas/__init__.py

@@ -0,0 +1,95 @@
+import os
+
+import cv2
+import numpy as np
+import torch
+from einops import rearrange
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from ..util import HWC3, resize_image
+from .api import MiDaSInference
+
+
+class MidasDetector:
+    def __init__(self, model):
+        self.model = model
+        
+    @classmethod
+    def from_pretrained(cls, pretrained_model_or_path, model_type="dpt_hybrid", filename=None, cache_dir=None):
+        if pretrained_model_or_path == "lllyasviel/ControlNet":
+            filename = filename or "annotator/ckpts/dpt_hybrid-midas-501f0c75.pt"
+        else:
+            filename = filename or "dpt_hybrid-midas-501f0c75.pt"
+
+        if os.path.isdir(pretrained_model_or_path):
+            model_path = os.path.join(pretrained_model_or_path, filename)
+        else:
+            model_path = hf_hub_download(pretrained_model_or_path, filename, cache_dir=cache_dir)
+
+        model = MiDaSInference(model_type=model_type, model_path=model_path)
+
+        return cls(model)
+        
+
+    def to(self, device):
+        self.model.to(device)
+        return self
+    
+    def __call__(self, input_image, a=np.pi * 2.0, bg_th=0.1, depth_and_normal=False, detect_resolution=512, image_resolution=512, output_type=None):
+        device = next(iter(self.model.parameters())).device
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+            output_type = output_type or "pil"
+        else:
+            output_type = output_type or "np"
+        
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+
+        assert input_image.ndim == 3
+        image_depth = input_image
+        with torch.no_grad():
+            image_depth = torch.from_numpy(image_depth).float()
+            image_depth = image_depth.to(device)
+            image_depth = image_depth / 127.5 - 1.0
+            image_depth = rearrange(image_depth, 'h w c -> 1 c h w')
+            depth = self.model(image_depth)[0]
+
+            depth_pt = depth.clone()
+            depth_pt -= torch.min(depth_pt)
+            depth_pt /= torch.max(depth_pt)
+            depth_pt = depth_pt.cpu().numpy()
+            depth_image = (depth_pt * 255.0).clip(0, 255).astype(np.uint8)
+
+            if depth_and_normal:
+                depth_np = depth.cpu().numpy()
+                x = cv2.Sobel(depth_np, cv2.CV_32F, 1, 0, ksize=3)
+                y = cv2.Sobel(depth_np, cv2.CV_32F, 0, 1, ksize=3)
+                z = np.ones_like(x) * a
+                x[depth_pt < bg_th] = 0
+                y[depth_pt < bg_th] = 0
+                normal = np.stack([x, y, z], axis=2)
+                normal /= np.sum(normal ** 2.0, axis=2, keepdims=True) ** 0.5
+                normal_image = (normal * 127.5 + 127.5).clip(0, 255).astype(np.uint8)[:, :, ::-1]
+        
+        depth_image = HWC3(depth_image)
+        if depth_and_normal:
+            normal_image = HWC3(normal_image)
+
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        depth_image = cv2.resize(depth_image, (W, H), interpolation=cv2.INTER_LINEAR)
+        if depth_and_normal:
+            normal_image = cv2.resize(normal_image, (W, H), interpolation=cv2.INTER_LINEAR)
+        
+        if output_type == "pil":
+            depth_image = Image.fromarray(depth_image)
+            if depth_and_normal:
+                normal_image = Image.fromarray(normal_image)
+        
+        if depth_and_normal:
+            return depth_image, normal_image
+        else:
+            return depth_image

negi/repos/controlnet_aux/src/controlnet_aux/midas/api.py

@@ -0,0 +1,169 @@
+# based on https://github.com/isl-org/MiDaS
+
+import cv2
+import os
+import torch
+import torch.nn as nn
+from torchvision.transforms import Compose
+
+from .midas.dpt_depth import DPTDepthModel
+from .midas.midas_net import MidasNet
+from .midas.midas_net_custom import MidasNet_small
+from .midas.transforms import Resize, NormalizeImage, PrepareForNet
+from ..util import annotator_ckpts_path
+
+
+ISL_PATHS = {
+    "dpt_large": os.path.join(annotator_ckpts_path, "dpt_large-midas-2f21e586.pt"),
+    "dpt_hybrid": os.path.join(annotator_ckpts_path, "dpt_hybrid-midas-501f0c75.pt"),
+    "midas_v21": "",
+    "midas_v21_small": "",
+}
+
+remote_model_path = "https://huggingface.co/lllyasviel/ControlNet/resolve/main/annotator/ckpts/dpt_hybrid-midas-501f0c75.pt"
+
+
+def disabled_train(self, mode=True):
+    """Overwrite model.train with this function to make sure train/eval mode
+    does not change anymore."""
+    return self
+
+
+def load_midas_transform(model_type):
+    # https://github.com/isl-org/MiDaS/blob/master/run.py
+    # load transform only
+    if model_type == "dpt_large":  # DPT-Large
+        net_w, net_h = 384, 384
+        resize_mode = "minimal"
+        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+
+    elif model_type == "dpt_hybrid":  # DPT-Hybrid
+        net_w, net_h = 384, 384
+        resize_mode = "minimal"
+        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+
+    elif model_type == "midas_v21":
+        net_w, net_h = 384, 384
+        resize_mode = "upper_bound"
+        normalization = NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+
+    elif model_type == "midas_v21_small":
+        net_w, net_h = 256, 256
+        resize_mode = "upper_bound"
+        normalization = NormalizeImage(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+
+    else:
+        assert False, f"model_type '{model_type}' not implemented, use: --model_type large"
+
+    transform = Compose(
+        [
+            Resize(
+                net_w,
+                net_h,
+                resize_target=None,
+                keep_aspect_ratio=True,
+                ensure_multiple_of=32,
+                resize_method=resize_mode,
+                image_interpolation_method=cv2.INTER_CUBIC,
+            ),
+            normalization,
+            PrepareForNet(),
+        ]
+    )
+
+    return transform
+
+
+def load_model(model_type, model_path=None):
+    # https://github.com/isl-org/MiDaS/blob/master/run.py
+    # load network
+    model_path = model_path or ISL_PATHS[model_type]
+    if model_type == "dpt_large":  # DPT-Large
+        model = DPTDepthModel(
+            path=model_path,
+            backbone="vitl16_384",
+            non_negative=True,
+        )
+        net_w, net_h = 384, 384
+        resize_mode = "minimal"
+        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+
+    elif model_type == "dpt_hybrid":  # DPT-Hybrid
+        if not os.path.exists(model_path):
+            from basicsr.utils.download_util import load_file_from_url
+            load_file_from_url(remote_model_path, model_dir=annotator_ckpts_path)
+
+        model = DPTDepthModel(
+            path=model_path,
+            backbone="vitb_rn50_384",
+            non_negative=True,
+        )
+        net_w, net_h = 384, 384
+        resize_mode = "minimal"
+        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+
+    elif model_type == "midas_v21":
+        model = MidasNet(model_path, non_negative=True)
+        net_w, net_h = 384, 384
+        resize_mode = "upper_bound"
+        normalization = NormalizeImage(
+            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
+        )
+
+    elif model_type == "midas_v21_small":
+        model = MidasNet_small(model_path, features=64, backbone="efficientnet_lite3", exportable=True,
+                               non_negative=True, blocks={'expand': True})
+        net_w, net_h = 256, 256
+        resize_mode = "upper_bound"
+        normalization = NormalizeImage(
+            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
+        )
+
+    else:
+        print(f"model_type '{model_type}' not implemented, use: --model_type large")
+        assert False
+
+    transform = Compose(
+        [
+            Resize(
+                net_w,
+                net_h,
+                resize_target=None,
+                keep_aspect_ratio=True,
+                ensure_multiple_of=32,
+                resize_method=resize_mode,
+                image_interpolation_method=cv2.INTER_CUBIC,
+            ),
+            normalization,
+            PrepareForNet(),
+        ]
+    )
+
+    return model.eval(), transform
+
+
+class MiDaSInference(nn.Module):
+    MODEL_TYPES_TORCH_HUB = [
+        "DPT_Large",
+        "DPT_Hybrid",
+        "MiDaS_small"
+    ]
+    MODEL_TYPES_ISL = [
+        "dpt_large",
+        "dpt_hybrid",
+        "midas_v21",
+        "midas_v21_small",
+    ]
+
+    def __init__(self, model_type, model_path):
+        super().__init__()
+        assert (model_type in self.MODEL_TYPES_ISL)
+        model, _ = load_model(model_type, model_path)
+        self.model = model
+        self.model.train = disabled_train
+
+    def forward(self, x):
+        with torch.no_grad():
+            prediction = self.model(x)
+        return prediction
+

negi/repos/controlnet_aux/src/controlnet_aux/midas/midas/base_model.py

@@ -0,0 +1,16 @@
+import torch
+
+
+class BaseModel(torch.nn.Module):
+    def load(self, path):
+        """Load model from file.
+
+        Args:
+            path (str): file path
+        """
+        parameters = torch.load(path, map_location=torch.device('cpu'))
+
+        if "optimizer" in parameters:
+            parameters = parameters["model"]
+
+        self.load_state_dict(parameters)

negi/repos/controlnet_aux/src/controlnet_aux/midas/midas/blocks.py

@@ -0,0 +1,342 @@
+import torch
+import torch.nn as nn
+
+from .vit import (
+    _make_pretrained_vitb_rn50_384,
+    _make_pretrained_vitl16_384,
+    _make_pretrained_vitb16_384,
+    forward_vit,
+)
+
+def _make_encoder(backbone, features, use_pretrained, groups=1, expand=False, exportable=True, hooks=None, use_vit_only=False, use_readout="ignore",):
+    if backbone == "vitl16_384":
+        pretrained = _make_pretrained_vitl16_384(
+            use_pretrained, hooks=hooks, use_readout=use_readout
+        )
+        scratch = _make_scratch(
+            [256, 512, 1024, 1024], features, groups=groups, expand=expand
+        )  # ViT-L/16 - 85.0% Top1 (backbone)
+    elif backbone == "vitb_rn50_384":
+        pretrained = _make_pretrained_vitb_rn50_384(
+            use_pretrained,
+            hooks=hooks,
+            use_vit_only=use_vit_only,
+            use_readout=use_readout,
+        )
+        scratch = _make_scratch(
+            [256, 512, 768, 768], features, groups=groups, expand=expand
+        )  # ViT-H/16 - 85.0% Top1 (backbone)
+    elif backbone == "vitb16_384":
+        pretrained = _make_pretrained_vitb16_384(
+            use_pretrained, hooks=hooks, use_readout=use_readout
+        )
+        scratch = _make_scratch(
+            [96, 192, 384, 768], features, groups=groups, expand=expand
+        )  # ViT-B/16 - 84.6% Top1 (backbone)
+    elif backbone == "resnext101_wsl":
+        pretrained = _make_pretrained_resnext101_wsl(use_pretrained)
+        scratch = _make_scratch([256, 512, 1024, 2048], features, groups=groups, expand=expand)     # efficientnet_lite3  
+    elif backbone == "efficientnet_lite3":
+        pretrained = _make_pretrained_efficientnet_lite3(use_pretrained, exportable=exportable)
+        scratch = _make_scratch([32, 48, 136, 384], features, groups=groups, expand=expand)  # efficientnet_lite3     
+    else:
+        print(f"Backbone '{backbone}' not implemented")
+        assert False
+        
+    return pretrained, scratch
+
+
+def _make_scratch(in_shape, out_shape, groups=1, expand=False):
+    scratch = nn.Module()
+
+    out_shape1 = out_shape
+    out_shape2 = out_shape
+    out_shape3 = out_shape
+    out_shape4 = out_shape
+    if expand==True:
+        out_shape1 = out_shape
+        out_shape2 = out_shape*2
+        out_shape3 = out_shape*4
+        out_shape4 = out_shape*8
+
+    scratch.layer1_rn = nn.Conv2d(
+        in_shape[0], out_shape1, kernel_size=3, stride=1, padding=1, bias=False, groups=groups
+    )
+    scratch.layer2_rn = nn.Conv2d(
+        in_shape[1], out_shape2, kernel_size=3, stride=1, padding=1, bias=False, groups=groups
+    )
+    scratch.layer3_rn = nn.Conv2d(
+        in_shape[2], out_shape3, kernel_size=3, stride=1, padding=1, bias=False, groups=groups
+    )
+    scratch.layer4_rn = nn.Conv2d(
+        in_shape[3], out_shape4, kernel_size=3, stride=1, padding=1, bias=False, groups=groups
+    )
+
+    return scratch
+
+
+def _make_pretrained_efficientnet_lite3(use_pretrained, exportable=False):
+    efficientnet = torch.hub.load(
+        "rwightman/gen-efficientnet-pytorch",
+        "tf_efficientnet_lite3",
+        pretrained=use_pretrained,
+        exportable=exportable
+    )
+    return _make_efficientnet_backbone(efficientnet)
+
+
+def _make_efficientnet_backbone(effnet):
+    pretrained = nn.Module()
+
+    pretrained.layer1 = nn.Sequential(
+        effnet.conv_stem, effnet.bn1, effnet.act1, *effnet.blocks[0:2]
+    )
+    pretrained.layer2 = nn.Sequential(*effnet.blocks[2:3])
+    pretrained.layer3 = nn.Sequential(*effnet.blocks[3:5])
+    pretrained.layer4 = nn.Sequential(*effnet.blocks[5:9])
+
+    return pretrained
+    
+
+def _make_resnet_backbone(resnet):
+    pretrained = nn.Module()
+    pretrained.layer1 = nn.Sequential(
+        resnet.conv1, resnet.bn1, resnet.relu, resnet.maxpool, resnet.layer1
+    )
+
+    pretrained.layer2 = resnet.layer2
+    pretrained.layer3 = resnet.layer3
+    pretrained.layer4 = resnet.layer4
+
+    return pretrained
+
+
+def _make_pretrained_resnext101_wsl(use_pretrained):
+    resnet = torch.hub.load("facebookresearch/WSL-Images", "resnext101_32x8d_wsl")
+    return _make_resnet_backbone(resnet)
+
+
+
+class Interpolate(nn.Module):
+    """Interpolation module.
+    """
+
+    def __init__(self, scale_factor, mode, align_corners=False):
+        """Init.
+
+        Args:
+            scale_factor (float): scaling
+            mode (str): interpolation mode
+        """
+        super(Interpolate, self).__init__()
+
+        self.interp = nn.functional.interpolate
+        self.scale_factor = scale_factor
+        self.mode = mode
+        self.align_corners = align_corners
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input
+
+        Returns:
+            tensor: interpolated data
+        """
+
+        x = self.interp(
+            x, scale_factor=self.scale_factor, mode=self.mode, align_corners=self.align_corners
+        )
+
+        return x
+
+
+class ResidualConvUnit(nn.Module):
+    """Residual convolution module.
+    """
+
+    def __init__(self, features):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super().__init__()
+
+        self.conv1 = nn.Conv2d(
+            features, features, kernel_size=3, stride=1, padding=1, bias=True
+        )
+
+        self.conv2 = nn.Conv2d(
+            features, features, kernel_size=3, stride=1, padding=1, bias=True
+        )
+
+        self.relu = nn.ReLU(inplace=True)
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input
+
+        Returns:
+            tensor: output
+        """
+        out = self.relu(x)
+        out = self.conv1(out)
+        out = self.relu(out)
+        out = self.conv2(out)
+
+        return out + x
+
+
+class FeatureFusionBlock(nn.Module):
+    """Feature fusion block.
+    """
+
+    def __init__(self, features):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super(FeatureFusionBlock, self).__init__()
+
+        self.resConfUnit1 = ResidualConvUnit(features)
+        self.resConfUnit2 = ResidualConvUnit(features)
+
+    def forward(self, *xs):
+        """Forward pass.
+
+        Returns:
+            tensor: output
+        """
+        output = xs[0]
+
+        if len(xs) == 2:
+            output += self.resConfUnit1(xs[1])
+
+        output = self.resConfUnit2(output)
+
+        output = nn.functional.interpolate(
+            output, scale_factor=2, mode="bilinear", align_corners=True
+        )
+
+        return output
+
+
+
+
+class ResidualConvUnit_custom(nn.Module):
+    """Residual convolution module.
+    """
+
+    def __init__(self, features, activation, bn):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super().__init__()
+
+        self.bn = bn
+
+        self.groups=1
+
+        self.conv1 = nn.Conv2d(
+            features, features, kernel_size=3, stride=1, padding=1, bias=True, groups=self.groups
+        )
+        
+        self.conv2 = nn.Conv2d(
+            features, features, kernel_size=3, stride=1, padding=1, bias=True, groups=self.groups
+        )
+
+        if self.bn==True:
+            self.bn1 = nn.BatchNorm2d(features)
+            self.bn2 = nn.BatchNorm2d(features)
+
+        self.activation = activation
+
+        self.skip_add = nn.quantized.FloatFunctional()
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input
+
+        Returns:
+            tensor: output
+        """
+        
+        out = self.activation(x)
+        out = self.conv1(out)
+        if self.bn==True:
+            out = self.bn1(out)
+       
+        out = self.activation(out)
+        out = self.conv2(out)
+        if self.bn==True:
+            out = self.bn2(out)
+
+        if self.groups > 1:
+            out = self.conv_merge(out)
+
+        return self.skip_add.add(out, x)
+
+        # return out + x
+
+
+class FeatureFusionBlock_custom(nn.Module):
+    """Feature fusion block.
+    """
+
+    def __init__(self, features, activation, deconv=False, bn=False, expand=False, align_corners=True):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super(FeatureFusionBlock_custom, self).__init__()
+
+        self.deconv = deconv
+        self.align_corners = align_corners
+
+        self.groups=1
+
+        self.expand = expand
+        out_features = features
+        if self.expand==True:
+            out_features = features//2
+        
+        self.out_conv = nn.Conv2d(features, out_features, kernel_size=1, stride=1, padding=0, bias=True, groups=1)
+
+        self.resConfUnit1 = ResidualConvUnit_custom(features, activation, bn)
+        self.resConfUnit2 = ResidualConvUnit_custom(features, activation, bn)
+        
+        self.skip_add = nn.quantized.FloatFunctional()
+
+    def forward(self, *xs):
+        """Forward pass.
+
+        Returns:
+            tensor: output
+        """
+        output = xs[0]
+
+        if len(xs) == 2:
+            res = self.resConfUnit1(xs[1])
+            output = self.skip_add.add(output, res)
+            # output += res
+
+        output = self.resConfUnit2(output)
+
+        output = nn.functional.interpolate(
+            output, scale_factor=2, mode="bilinear", align_corners=self.align_corners
+        )
+
+        output = self.out_conv(output)
+
+        return output
+

negi/repos/controlnet_aux/src/controlnet_aux/midas/midas/dpt_depth.py

@@ -0,0 +1,109 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .base_model import BaseModel
+from .blocks import (
+    FeatureFusionBlock,
+    FeatureFusionBlock_custom,
+    Interpolate,
+    _make_encoder,
+    forward_vit,
+)
+
+
+def _make_fusion_block(features, use_bn):
+    return FeatureFusionBlock_custom(
+        features,
+        nn.ReLU(False),
+        deconv=False,
+        bn=use_bn,
+        expand=False,
+        align_corners=True,
+    )
+
+
+class DPT(BaseModel):
+    def __init__(
+        self,
+        head,
+        features=256,
+        backbone="vitb_rn50_384",
+        readout="project",
+        channels_last=False,
+        use_bn=False,
+    ):
+
+        super(DPT, self).__init__()
+
+        self.channels_last = channels_last
+
+        hooks = {
+            "vitb_rn50_384": [0, 1, 8, 11],
+            "vitb16_384": [2, 5, 8, 11],
+            "vitl16_384": [5, 11, 17, 23],
+        }
+
+        # Instantiate backbone and reassemble blocks
+        self.pretrained, self.scratch = _make_encoder(
+            backbone,
+            features,
+            False, # Set to true of you want to train from scratch, uses ImageNet weights
+            groups=1,
+            expand=False,
+            exportable=False,
+            hooks=hooks[backbone],
+            use_readout=readout,
+        )
+
+        self.scratch.refinenet1 = _make_fusion_block(features, use_bn)
+        self.scratch.refinenet2 = _make_fusion_block(features, use_bn)
+        self.scratch.refinenet3 = _make_fusion_block(features, use_bn)
+        self.scratch.refinenet4 = _make_fusion_block(features, use_bn)
+
+        self.scratch.output_conv = head
+
+
+    def forward(self, x):
+        if self.channels_last == True:
+            x.contiguous(memory_format=torch.channels_last)
+
+        layer_1, layer_2, layer_3, layer_4 = forward_vit(self.pretrained, x)
+
+        layer_1_rn = self.scratch.layer1_rn(layer_1)
+        layer_2_rn = self.scratch.layer2_rn(layer_2)
+        layer_3_rn = self.scratch.layer3_rn(layer_3)
+        layer_4_rn = self.scratch.layer4_rn(layer_4)
+
+        path_4 = self.scratch.refinenet4(layer_4_rn)
+        path_3 = self.scratch.refinenet3(path_4, layer_3_rn)
+        path_2 = self.scratch.refinenet2(path_3, layer_2_rn)
+        path_1 = self.scratch.refinenet1(path_2, layer_1_rn)
+
+        out = self.scratch.output_conv(path_1)
+
+        return out
+
+
+class DPTDepthModel(DPT):
+    def __init__(self, path=None, non_negative=True, **kwargs):
+        features = kwargs["features"] if "features" in kwargs else 256
+
+        head = nn.Sequential(
+            nn.Conv2d(features, features // 2, kernel_size=3, stride=1, padding=1),
+            Interpolate(scale_factor=2, mode="bilinear", align_corners=True),
+            nn.Conv2d(features // 2, 32, kernel_size=3, stride=1, padding=1),
+            nn.ReLU(True),
+            nn.Conv2d(32, 1, kernel_size=1, stride=1, padding=0),
+            nn.ReLU(True) if non_negative else nn.Identity(),
+            nn.Identity(),
+        )
+
+        super().__init__(head, **kwargs)
+
+        if path is not None:
+           self.load(path)
+
+    def forward(self, x):
+        return super().forward(x).squeeze(dim=1)
+

negi/repos/controlnet_aux/src/controlnet_aux/midas/midas/midas_net.py

@@ -0,0 +1,76 @@
+"""MidashNet: Network for monocular depth estimation trained by mixing several datasets.
+This file contains code that is adapted from
+https://github.com/thomasjpfan/pytorch_refinenet/blob/master/pytorch_refinenet/refinenet/refinenet_4cascade.py
+"""
+import torch
+import torch.nn as nn
+
+from .base_model import BaseModel
+from .blocks import FeatureFusionBlock, Interpolate, _make_encoder
+
+
+class MidasNet(BaseModel):
+    """Network for monocular depth estimation.
+    """
+
+    def __init__(self, path=None, features=256, non_negative=True):
+        """Init.
+
+        Args:
+            path (str, optional): Path to saved model. Defaults to None.
+            features (int, optional): Number of features. Defaults to 256.
+            backbone (str, optional): Backbone network for encoder. Defaults to resnet50
+        """
+        print("Loading weights: ", path)
+
+        super(MidasNet, self).__init__()
+
+        use_pretrained = False if path is None else True
+
+        self.pretrained, self.scratch = _make_encoder(backbone="resnext101_wsl", features=features, use_pretrained=use_pretrained)
+
+        self.scratch.refinenet4 = FeatureFusionBlock(features)
+        self.scratch.refinenet3 = FeatureFusionBlock(features)
+        self.scratch.refinenet2 = FeatureFusionBlock(features)
+        self.scratch.refinenet1 = FeatureFusionBlock(features)
+
+        self.scratch.output_conv = nn.Sequential(
+            nn.Conv2d(features, 128, kernel_size=3, stride=1, padding=1),
+            Interpolate(scale_factor=2, mode="bilinear"),
+            nn.Conv2d(128, 32, kernel_size=3, stride=1, padding=1),
+            nn.ReLU(True),
+            nn.Conv2d(32, 1, kernel_size=1, stride=1, padding=0),
+            nn.ReLU(True) if non_negative else nn.Identity(),
+        )
+
+        if path:
+            self.load(path)
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input data (image)
+
+        Returns:
+            tensor: depth
+        """
+
+        layer_1 = self.pretrained.layer1(x)
+        layer_2 = self.pretrained.layer2(layer_1)
+        layer_3 = self.pretrained.layer3(layer_2)
+        layer_4 = self.pretrained.layer4(layer_3)
+
+        layer_1_rn = self.scratch.layer1_rn(layer_1)
+        layer_2_rn = self.scratch.layer2_rn(layer_2)
+        layer_3_rn = self.scratch.layer3_rn(layer_3)
+        layer_4_rn = self.scratch.layer4_rn(layer_4)
+
+        path_4 = self.scratch.refinenet4(layer_4_rn)
+        path_3 = self.scratch.refinenet3(path_4, layer_3_rn)
+        path_2 = self.scratch.refinenet2(path_3, layer_2_rn)
+        path_1 = self.scratch.refinenet1(path_2, layer_1_rn)
+
+        out = self.scratch.output_conv(path_1)
+
+        return torch.squeeze(out, dim=1)

negi/repos/controlnet_aux/src/controlnet_aux/midas/midas/midas_net_custom.py

@@ -0,0 +1,128 @@
+"""MidashNet: Network for monocular depth estimation trained by mixing several datasets.
+This file contains code that is adapted from
+https://github.com/thomasjpfan/pytorch_refinenet/blob/master/pytorch_refinenet/refinenet/refinenet_4cascade.py
+"""
+import torch
+import torch.nn as nn
+
+from .base_model import BaseModel
+from .blocks import FeatureFusionBlock, FeatureFusionBlock_custom, Interpolate, _make_encoder
+
+
+class MidasNet_small(BaseModel):
+    """Network for monocular depth estimation.
+    """
+
+    def __init__(self, path=None, features=64, backbone="efficientnet_lite3", non_negative=True, exportable=True, channels_last=False, align_corners=True,
+        blocks={'expand': True}):
+        """Init.
+
+        Args:
+            path (str, optional): Path to saved model. Defaults to None.
+            features (int, optional): Number of features. Defaults to 256.
+            backbone (str, optional): Backbone network for encoder. Defaults to resnet50
+        """
+        print("Loading weights: ", path)
+
+        super(MidasNet_small, self).__init__()
+
+        use_pretrained = False if path else True
+                
+        self.channels_last = channels_last
+        self.blocks = blocks
+        self.backbone = backbone
+
+        self.groups = 1
+
+        features1=features
+        features2=features
+        features3=features
+        features4=features
+        self.expand = False
+        if "expand" in self.blocks and self.blocks['expand'] == True:
+            self.expand = True
+            features1=features
+            features2=features*2
+            features3=features*4
+            features4=features*8
+
+        self.pretrained, self.scratch = _make_encoder(self.backbone, features, use_pretrained, groups=self.groups, expand=self.expand, exportable=exportable)
+  
+        self.scratch.activation = nn.ReLU(False)    
+
+        self.scratch.refinenet4 = FeatureFusionBlock_custom(features4, self.scratch.activation, deconv=False, bn=False, expand=self.expand, align_corners=align_corners)
+        self.scratch.refinenet3 = FeatureFusionBlock_custom(features3, self.scratch.activation, deconv=False, bn=False, expand=self.expand, align_corners=align_corners)
+        self.scratch.refinenet2 = FeatureFusionBlock_custom(features2, self.scratch.activation, deconv=False, bn=False, expand=self.expand, align_corners=align_corners)
+        self.scratch.refinenet1 = FeatureFusionBlock_custom(features1, self.scratch.activation, deconv=False, bn=False, align_corners=align_corners)
+
+        
+        self.scratch.output_conv = nn.Sequential(
+            nn.Conv2d(features, features//2, kernel_size=3, stride=1, padding=1, groups=self.groups),
+            Interpolate(scale_factor=2, mode="bilinear"),
+            nn.Conv2d(features//2, 32, kernel_size=3, stride=1, padding=1),
+            self.scratch.activation,
+            nn.Conv2d(32, 1, kernel_size=1, stride=1, padding=0),
+            nn.ReLU(True) if non_negative else nn.Identity(),
+            nn.Identity(),
+        )
+        
+        if path:
+            self.load(path)
+
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input data (image)
+
+        Returns:
+            tensor: depth
+        """
+        if self.channels_last==True:
+            print("self.channels_last = ", self.channels_last)
+            x.contiguous(memory_format=torch.channels_last)
+
+
+        layer_1 = self.pretrained.layer1(x)
+        layer_2 = self.pretrained.layer2(layer_1)
+        layer_3 = self.pretrained.layer3(layer_2)
+        layer_4 = self.pretrained.layer4(layer_3)
+        
+        layer_1_rn = self.scratch.layer1_rn(layer_1)
+        layer_2_rn = self.scratch.layer2_rn(layer_2)
+        layer_3_rn = self.scratch.layer3_rn(layer_3)
+        layer_4_rn = self.scratch.layer4_rn(layer_4)
+
+
+        path_4 = self.scratch.refinenet4(layer_4_rn)
+        path_3 = self.scratch.refinenet3(path_4, layer_3_rn)
+        path_2 = self.scratch.refinenet2(path_3, layer_2_rn)
+        path_1 = self.scratch.refinenet1(path_2, layer_1_rn)
+        
+        out = self.scratch.output_conv(path_1)
+
+        return torch.squeeze(out, dim=1)
+
+
+
+def fuse_model(m):
+    prev_previous_type = nn.Identity()
+    prev_previous_name = ''
+    previous_type = nn.Identity()
+    previous_name = ''
+    for name, module in m.named_modules():
+        if prev_previous_type == nn.Conv2d and previous_type == nn.BatchNorm2d and type(module) == nn.ReLU:
+            # print("FUSED ", prev_previous_name, previous_name, name)
+            torch.quantization.fuse_modules(m, [prev_previous_name, previous_name, name], inplace=True)
+        elif prev_previous_type == nn.Conv2d and previous_type == nn.BatchNorm2d:
+            # print("FUSED ", prev_previous_name, previous_name)
+            torch.quantization.fuse_modules(m, [prev_previous_name, previous_name], inplace=True)
+        # elif previous_type == nn.Conv2d and type(module) == nn.ReLU:
+        #    print("FUSED ", previous_name, name)
+        #    torch.quantization.fuse_modules(m, [previous_name, name], inplace=True)
+
+        prev_previous_type = previous_type
+        prev_previous_name = previous_name
+        previous_type = type(module)
+        previous_name = name

negi/repos/controlnet_aux/src/controlnet_aux/midas/midas/transforms.py

@@ -0,0 +1,234 @@
+import numpy as np
+import cv2
+import math
+
+
+def apply_min_size(sample, size, image_interpolation_method=cv2.INTER_AREA):
+    """Rezise the sample to ensure the given size. Keeps aspect ratio.
+
+    Args:
+        sample (dict): sample
+        size (tuple): image size
+
+    Returns:
+        tuple: new size
+    """
+    shape = list(sample["disparity"].shape)
+
+    if shape[0] >= size[0] and shape[1] >= size[1]:
+        return sample
+
+    scale = [0, 0]
+    scale[0] = size[0] / shape[0]
+    scale[1] = size[1] / shape[1]
+
+    scale = max(scale)
+
+    shape[0] = math.ceil(scale * shape[0])
+    shape[1] = math.ceil(scale * shape[1])
+
+    # resize
+    sample["image"] = cv2.resize(
+        sample["image"], tuple(shape[::-1]), interpolation=image_interpolation_method
+    )
+
+    sample["disparity"] = cv2.resize(
+        sample["disparity"], tuple(shape[::-1]), interpolation=cv2.INTER_NEAREST
+    )
+    sample["mask"] = cv2.resize(
+        sample["mask"].astype(np.float32),
+        tuple(shape[::-1]),
+        interpolation=cv2.INTER_NEAREST,
+    )
+    sample["mask"] = sample["mask"].astype(bool)
+
+    return tuple(shape)
+
+
+class Resize(object):
+    """Resize sample to given size (width, height).
+    """
+
+    def __init__(
+        self,
+        width,
+        height,
+        resize_target=True,
+        keep_aspect_ratio=False,
+        ensure_multiple_of=1,
+        resize_method="lower_bound",
+        image_interpolation_method=cv2.INTER_AREA,
+    ):
+        """Init.
+
+        Args:
+            width (int): desired output width
+            height (int): desired output height
+            resize_target (bool, optional):
+                True: Resize the full sample (image, mask, target).
+                False: Resize image only.
+                Defaults to True.
+            keep_aspect_ratio (bool, optional):
+                True: Keep the aspect ratio of the input sample.
+                Output sample might not have the given width and height, and
+                resize behaviour depends on the parameter 'resize_method'.
+                Defaults to False.
+            ensure_multiple_of (int, optional):
+                Output width and height is constrained to be multiple of this parameter.
+                Defaults to 1.
+            resize_method (str, optional):
+                "lower_bound": Output will be at least as large as the given size.
+                "upper_bound": Output will be at max as large as the given size. (Output size might be smaller than given size.)
+                "minimal": Scale as least as possible.  (Output size might be smaller than given size.)
+                Defaults to "lower_bound".
+        """
+        self.__width = width
+        self.__height = height
+
+        self.__resize_target = resize_target
+        self.__keep_aspect_ratio = keep_aspect_ratio
+        self.__multiple_of = ensure_multiple_of
+        self.__resize_method = resize_method
+        self.__image_interpolation_method = image_interpolation_method
+
+    def constrain_to_multiple_of(self, x, min_val=0, max_val=None):
+        y = (np.round(x / self.__multiple_of) * self.__multiple_of).astype(int)
+
+        if max_val is not None and y > max_val:
+            y = (np.floor(x / self.__multiple_of) * self.__multiple_of).astype(int)
+
+        if y < min_val:
+            y = (np.ceil(x / self.__multiple_of) * self.__multiple_of).astype(int)
+
+        return y
+
+    def get_size(self, width, height):
+        # determine new height and width
+        scale_height = self.__height / height
+        scale_width = self.__width / width
+
+        if self.__keep_aspect_ratio:
+            if self.__resize_method == "lower_bound":
+                # scale such that output size is lower bound
+                if scale_width > scale_height:
+                    # fit width
+                    scale_height = scale_width
+                else:
+                    # fit height
+                    scale_width = scale_height
+            elif self.__resize_method == "upper_bound":
+                # scale such that output size is upper bound
+                if scale_width < scale_height:
+                    # fit width
+                    scale_height = scale_width
+                else:
+                    # fit height
+                    scale_width = scale_height
+            elif self.__resize_method == "minimal":
+                # scale as least as possbile
+                if abs(1 - scale_width) < abs(1 - scale_height):
+                    # fit width
+                    scale_height = scale_width
+                else:
+                    # fit height
+                    scale_width = scale_height
+            else:
+                raise ValueError(
+                    f"resize_method {self.__resize_method} not implemented"
+                )
+
+        if self.__resize_method == "lower_bound":
+            new_height = self.constrain_to_multiple_of(
+                scale_height * height, min_val=self.__height
+            )
+            new_width = self.constrain_to_multiple_of(
+                scale_width * width, min_val=self.__width
+            )
+        elif self.__resize_method == "upper_bound":
+            new_height = self.constrain_to_multiple_of(
+                scale_height * height, max_val=self.__height
+            )
+            new_width = self.constrain_to_multiple_of(
+                scale_width * width, max_val=self.__width
+            )
+        elif self.__resize_method == "minimal":
+            new_height = self.constrain_to_multiple_of(scale_height * height)
+            new_width = self.constrain_to_multiple_of(scale_width * width)
+        else:
+            raise ValueError(f"resize_method {self.__resize_method} not implemented")
+
+        return (new_width, new_height)
+
+    def __call__(self, sample):
+        width, height = self.get_size(
+            sample["image"].shape[1], sample["image"].shape[0]
+        )
+
+        # resize sample
+        sample["image"] = cv2.resize(
+            sample["image"],
+            (width, height),
+            interpolation=self.__image_interpolation_method,
+        )
+
+        if self.__resize_target:
+            if "disparity" in sample:
+                sample["disparity"] = cv2.resize(
+                    sample["disparity"],
+                    (width, height),
+                    interpolation=cv2.INTER_NEAREST,
+                )
+
+            if "depth" in sample:
+                sample["depth"] = cv2.resize(
+                    sample["depth"], (width, height), interpolation=cv2.INTER_NEAREST
+                )
+
+            sample["mask"] = cv2.resize(
+                sample["mask"].astype(np.float32),
+                (width, height),
+                interpolation=cv2.INTER_NEAREST,
+            )
+            sample["mask"] = sample["mask"].astype(bool)
+
+        return sample
+
+
+class NormalizeImage(object):
+    """Normlize image by given mean and std.
+    """
+
+    def __init__(self, mean, std):
+        self.__mean = mean
+        self.__std = std
+
+    def __call__(self, sample):
+        sample["image"] = (sample["image"] - self.__mean) / self.__std
+
+        return sample
+
+
+class PrepareForNet(object):
+    """Prepare sample for usage as network input.
+    """
+
+    def __init__(self):
+        pass
+
+    def __call__(self, sample):
+        image = np.transpose(sample["image"], (2, 0, 1))
+        sample["image"] = np.ascontiguousarray(image).astype(np.float32)
+
+        if "mask" in sample:
+            sample["mask"] = sample["mask"].astype(np.float32)
+            sample["mask"] = np.ascontiguousarray(sample["mask"])
+
+        if "disparity" in sample:
+            disparity = sample["disparity"].astype(np.float32)
+            sample["disparity"] = np.ascontiguousarray(disparity)
+
+        if "depth" in sample:
+            depth = sample["depth"].astype(np.float32)
+            sample["depth"] = np.ascontiguousarray(depth)
+
+        return sample

negi/repos/controlnet_aux/src/controlnet_aux/midas/midas/vit.py

@@ -0,0 +1,491 @@
+import torch
+import torch.nn as nn
+import timm
+import types
+import math
+import torch.nn.functional as F
+
+
+class Slice(nn.Module):
+    def __init__(self, start_index=1):
+        super(Slice, self).__init__()
+        self.start_index = start_index
+
+    def forward(self, x):
+        return x[:, self.start_index :]
+
+
+class AddReadout(nn.Module):
+    def __init__(self, start_index=1):
+        super(AddReadout, self).__init__()
+        self.start_index = start_index
+
+    def forward(self, x):
+        if self.start_index == 2:
+            readout = (x[:, 0] + x[:, 1]) / 2
+        else:
+            readout = x[:, 0]
+        return x[:, self.start_index :] + readout.unsqueeze(1)
+
+
+class ProjectReadout(nn.Module):
+    def __init__(self, in_features, start_index=1):
+        super(ProjectReadout, self).__init__()
+        self.start_index = start_index
+
+        self.project = nn.Sequential(nn.Linear(2 * in_features, in_features), nn.GELU())
+
+    def forward(self, x):
+        readout = x[:, 0].unsqueeze(1).expand_as(x[:, self.start_index :])
+        features = torch.cat((x[:, self.start_index :], readout), -1)
+
+        return self.project(features)
+
+
+class Transpose(nn.Module):
+    def __init__(self, dim0, dim1):
+        super(Transpose, self).__init__()
+        self.dim0 = dim0
+        self.dim1 = dim1
+
+    def forward(self, x):
+        x = x.transpose(self.dim0, self.dim1)
+        return x
+
+
+def forward_vit(pretrained, x):
+    b, c, h, w = x.shape
+
+    glob = pretrained.model.forward_flex(x)
+
+    layer_1 = pretrained.activations["1"]
+    layer_2 = pretrained.activations["2"]
+    layer_3 = pretrained.activations["3"]
+    layer_4 = pretrained.activations["4"]
+
+    layer_1 = pretrained.act_postprocess1[0:2](layer_1)
+    layer_2 = pretrained.act_postprocess2[0:2](layer_2)
+    layer_3 = pretrained.act_postprocess3[0:2](layer_3)
+    layer_4 = pretrained.act_postprocess4[0:2](layer_4)
+
+    unflatten = nn.Sequential(
+        nn.Unflatten(
+            2,
+            torch.Size(
+                [
+                    h // pretrained.model.patch_size[1],
+                    w // pretrained.model.patch_size[0],
+                ]
+            ),
+        )
+    )
+
+    if layer_1.ndim == 3:
+        layer_1 = unflatten(layer_1)
+    if layer_2.ndim == 3:
+        layer_2 = unflatten(layer_2)
+    if layer_3.ndim == 3:
+        layer_3 = unflatten(layer_3)
+    if layer_4.ndim == 3:
+        layer_4 = unflatten(layer_4)
+
+    layer_1 = pretrained.act_postprocess1[3 : len(pretrained.act_postprocess1)](layer_1)
+    layer_2 = pretrained.act_postprocess2[3 : len(pretrained.act_postprocess2)](layer_2)
+    layer_3 = pretrained.act_postprocess3[3 : len(pretrained.act_postprocess3)](layer_3)
+    layer_4 = pretrained.act_postprocess4[3 : len(pretrained.act_postprocess4)](layer_4)
+
+    return layer_1, layer_2, layer_3, layer_4
+
+
+def _resize_pos_embed(self, posemb, gs_h, gs_w):
+    posemb_tok, posemb_grid = (
+        posemb[:, : self.start_index],
+        posemb[0, self.start_index :],
+    )
+
+    gs_old = int(math.sqrt(len(posemb_grid)))
+
+    posemb_grid = posemb_grid.reshape(1, gs_old, gs_old, -1).permute(0, 3, 1, 2)
+    posemb_grid = F.interpolate(posemb_grid, size=(gs_h, gs_w), mode="bilinear")
+    posemb_grid = posemb_grid.permute(0, 2, 3, 1).reshape(1, gs_h * gs_w, -1)
+
+    posemb = torch.cat([posemb_tok, posemb_grid], dim=1)
+
+    return posemb
+
+
+def forward_flex(self, x):
+    b, c, h, w = x.shape
+
+    pos_embed = self._resize_pos_embed(
+        self.pos_embed, h // self.patch_size[1], w // self.patch_size[0]
+    )
+
+    B = x.shape[0]
+
+    if hasattr(self.patch_embed, "backbone"):
+        x = self.patch_embed.backbone(x)
+        if isinstance(x, (list, tuple)):
+            x = x[-1]  # last feature if backbone outputs list/tuple of features
+
+    x = self.patch_embed.proj(x).flatten(2).transpose(1, 2)
+
+    if getattr(self, "dist_token", None) is not None:
+        cls_tokens = self.cls_token.expand(
+            B, -1, -1
+        )  # stole cls_tokens impl from Phil Wang, thanks
+        dist_token = self.dist_token.expand(B, -1, -1)
+        x = torch.cat((cls_tokens, dist_token, x), dim=1)
+    else:
+        cls_tokens = self.cls_token.expand(
+            B, -1, -1
+        )  # stole cls_tokens impl from Phil Wang, thanks
+        x = torch.cat((cls_tokens, x), dim=1)
+
+    x = x + pos_embed
+    x = self.pos_drop(x)
+
+    for blk in self.blocks:
+        x = blk(x)
+
+    x = self.norm(x)
+
+    return x
+
+
+activations = {}
+
+
+def get_activation(name):
+    def hook(model, input, output):
+        activations[name] = output
+
+    return hook
+
+
+def get_readout_oper(vit_features, features, use_readout, start_index=1):
+    if use_readout == "ignore":
+        readout_oper = [Slice(start_index)] * len(features)
+    elif use_readout == "add":
+        readout_oper = [AddReadout(start_index)] * len(features)
+    elif use_readout == "project":
+        readout_oper = [
+            ProjectReadout(vit_features, start_index) for out_feat in features
+        ]
+    else:
+        assert (
+            False
+        ), "wrong operation for readout token, use_readout can be 'ignore', 'add', or 'project'"
+
+    return readout_oper
+
+
+def _make_vit_b16_backbone(
+    model,
+    features=[96, 192, 384, 768],
+    size=[384, 384],
+    hooks=[2, 5, 8, 11],
+    vit_features=768,
+    use_readout="ignore",
+    start_index=1,
+):
+    pretrained = nn.Module()
+
+    pretrained.model = model
+    pretrained.model.blocks[hooks[0]].register_forward_hook(get_activation("1"))
+    pretrained.model.blocks[hooks[1]].register_forward_hook(get_activation("2"))
+    pretrained.model.blocks[hooks[2]].register_forward_hook(get_activation("3"))
+    pretrained.model.blocks[hooks[3]].register_forward_hook(get_activation("4"))
+
+    pretrained.activations = activations
+
+    readout_oper = get_readout_oper(vit_features, features, use_readout, start_index)
+
+    # 32, 48, 136, 384
+    pretrained.act_postprocess1 = nn.Sequential(
+        readout_oper[0],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[0],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+        nn.ConvTranspose2d(
+            in_channels=features[0],
+            out_channels=features[0],
+            kernel_size=4,
+            stride=4,
+            padding=0,
+            bias=True,
+            dilation=1,
+            groups=1,
+        ),
+    )
+
+    pretrained.act_postprocess2 = nn.Sequential(
+        readout_oper[1],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[1],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+        nn.ConvTranspose2d(
+            in_channels=features[1],
+            out_channels=features[1],
+            kernel_size=2,
+            stride=2,
+            padding=0,
+            bias=True,
+            dilation=1,
+            groups=1,
+        ),
+    )
+
+    pretrained.act_postprocess3 = nn.Sequential(
+        readout_oper[2],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[2],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+    )
+
+    pretrained.act_postprocess4 = nn.Sequential(
+        readout_oper[3],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[3],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+        nn.Conv2d(
+            in_channels=features[3],
+            out_channels=features[3],
+            kernel_size=3,
+            stride=2,
+            padding=1,
+        ),
+    )
+
+    pretrained.model.start_index = start_index
+    pretrained.model.patch_size = [16, 16]
+
+    # We inject this function into the VisionTransformer instances so that
+    # we can use it with interpolated position embeddings without modifying the library source.
+    pretrained.model.forward_flex = types.MethodType(forward_flex, pretrained.model)
+    pretrained.model._resize_pos_embed = types.MethodType(
+        _resize_pos_embed, pretrained.model
+    )
+
+    return pretrained
+
+
+def _make_pretrained_vitl16_384(pretrained, use_readout="ignore", hooks=None):
+    model = timm.create_model("vit_large_patch16_384", pretrained=pretrained)
+
+    hooks = [5, 11, 17, 23] if hooks == None else hooks
+    return _make_vit_b16_backbone(
+        model,
+        features=[256, 512, 1024, 1024],
+        hooks=hooks,
+        vit_features=1024,
+        use_readout=use_readout,
+    )
+
+
+def _make_pretrained_vitb16_384(pretrained, use_readout="ignore", hooks=None):
+    model = timm.create_model("vit_base_patch16_384", pretrained=pretrained)
+
+    hooks = [2, 5, 8, 11] if hooks == None else hooks
+    return _make_vit_b16_backbone(
+        model, features=[96, 192, 384, 768], hooks=hooks, use_readout=use_readout
+    )
+
+
+def _make_pretrained_deitb16_384(pretrained, use_readout="ignore", hooks=None):
+    model = timm.create_model("vit_deit_base_patch16_384", pretrained=pretrained)
+
+    hooks = [2, 5, 8, 11] if hooks == None else hooks
+    return _make_vit_b16_backbone(
+        model, features=[96, 192, 384, 768], hooks=hooks, use_readout=use_readout
+    )
+
+
+def _make_pretrained_deitb16_distil_384(pretrained, use_readout="ignore", hooks=None):
+    model = timm.create_model(
+        "vit_deit_base_distilled_patch16_384", pretrained=pretrained
+    )
+
+    hooks = [2, 5, 8, 11] if hooks == None else hooks
+    return _make_vit_b16_backbone(
+        model,
+        features=[96, 192, 384, 768],
+        hooks=hooks,
+        use_readout=use_readout,
+        start_index=2,
+    )
+
+
+def _make_vit_b_rn50_backbone(
+    model,
+    features=[256, 512, 768, 768],
+    size=[384, 384],
+    hooks=[0, 1, 8, 11],
+    vit_features=768,
+    use_vit_only=False,
+    use_readout="ignore",
+    start_index=1,
+):
+    pretrained = nn.Module()
+
+    pretrained.model = model
+
+    if use_vit_only == True:
+        pretrained.model.blocks[hooks[0]].register_forward_hook(get_activation("1"))
+        pretrained.model.blocks[hooks[1]].register_forward_hook(get_activation("2"))
+    else:
+        pretrained.model.patch_embed.backbone.stages[0].register_forward_hook(
+            get_activation("1")
+        )
+        pretrained.model.patch_embed.backbone.stages[1].register_forward_hook(
+            get_activation("2")
+        )
+
+    pretrained.model.blocks[hooks[2]].register_forward_hook(get_activation("3"))
+    pretrained.model.blocks[hooks[3]].register_forward_hook(get_activation("4"))
+
+    pretrained.activations = activations
+
+    readout_oper = get_readout_oper(vit_features, features, use_readout, start_index)
+
+    if use_vit_only == True:
+        pretrained.act_postprocess1 = nn.Sequential(
+            readout_oper[0],
+            Transpose(1, 2),
+            nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+            nn.Conv2d(
+                in_channels=vit_features,
+                out_channels=features[0],
+                kernel_size=1,
+                stride=1,
+                padding=0,
+            ),
+            nn.ConvTranspose2d(
+                in_channels=features[0],
+                out_channels=features[0],
+                kernel_size=4,
+                stride=4,
+                padding=0,
+                bias=True,
+                dilation=1,
+                groups=1,
+            ),
+        )
+
+        pretrained.act_postprocess2 = nn.Sequential(
+            readout_oper[1],
+            Transpose(1, 2),
+            nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+            nn.Conv2d(
+                in_channels=vit_features,
+                out_channels=features[1],
+                kernel_size=1,
+                stride=1,
+                padding=0,
+            ),
+            nn.ConvTranspose2d(
+                in_channels=features[1],
+                out_channels=features[1],
+                kernel_size=2,
+                stride=2,
+                padding=0,
+                bias=True,
+                dilation=1,
+                groups=1,
+            ),
+        )
+    else:
+        pretrained.act_postprocess1 = nn.Sequential(
+            nn.Identity(), nn.Identity(), nn.Identity()
+        )
+        pretrained.act_postprocess2 = nn.Sequential(
+            nn.Identity(), nn.Identity(), nn.Identity()
+        )
+
+    pretrained.act_postprocess3 = nn.Sequential(
+        readout_oper[2],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[2],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+    )
+
+    pretrained.act_postprocess4 = nn.Sequential(
+        readout_oper[3],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[3],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+        nn.Conv2d(
+            in_channels=features[3],
+            out_channels=features[3],
+            kernel_size=3,
+            stride=2,
+            padding=1,
+        ),
+    )
+
+    pretrained.model.start_index = start_index
+    pretrained.model.patch_size = [16, 16]
+
+    # We inject this function into the VisionTransformer instances so that
+    # we can use it with interpolated position embeddings without modifying the library source.
+    pretrained.model.forward_flex = types.MethodType(forward_flex, pretrained.model)
+
+    # We inject this function into the VisionTransformer instances so that
+    # we can use it with interpolated position embeddings without modifying the library source.
+    pretrained.model._resize_pos_embed = types.MethodType(
+        _resize_pos_embed, pretrained.model
+    )
+
+    return pretrained
+
+
+def _make_pretrained_vitb_rn50_384(
+    pretrained, use_readout="ignore", hooks=None, use_vit_only=False
+):
+    model = timm.create_model("vit_base_resnet50_384", pretrained=pretrained)
+
+    hooks = [0, 1, 8, 11] if hooks == None else hooks
+    return _make_vit_b_rn50_backbone(
+        model,
+        features=[256, 512, 768, 768],
+        size=[384, 384],
+        hooks=hooks,
+        use_vit_only=use_vit_only,
+        use_readout=use_readout,
+    )

negi/repos/controlnet_aux/src/controlnet_aux/midas/utils.py

@@ -0,0 +1,189 @@
+"""Utils for monoDepth."""
+import sys
+import re
+import numpy as np
+import cv2
+import torch
+
+
+def read_pfm(path):
+    """Read pfm file.
+
+    Args:
+        path (str): path to file
+
+    Returns:
+        tuple: (data, scale)
+    """
+    with open(path, "rb") as file:
+
+        color = None
+        width = None
+        height = None
+        scale = None
+        endian = None
+
+        header = file.readline().rstrip()
+        if header.decode("ascii") == "PF":
+            color = True
+        elif header.decode("ascii") == "Pf":
+            color = False
+        else:
+            raise Exception("Not a PFM file: " + path)
+
+        dim_match = re.match(r"^(\d+)\s(\d+)\s$", file.readline().decode("ascii"))
+        if dim_match:
+            width, height = list(map(int, dim_match.groups()))
+        else:
+            raise Exception("Malformed PFM header.")
+
+        scale = float(file.readline().decode("ascii").rstrip())
+        if scale < 0:
+            # little-endian
+            endian = "<"
+            scale = -scale
+        else:
+            # big-endian
+            endian = ">"
+
+        data = np.fromfile(file, endian + "f")
+        shape = (height, width, 3) if color else (height, width)
+
+        data = np.reshape(data, shape)
+        data = np.flipud(data)
+
+        return data, scale
+
+
+def write_pfm(path, image, scale=1):
+    """Write pfm file.
+
+    Args:
+        path (str): pathto file
+        image (array): data
+        scale (int, optional): Scale. Defaults to 1.
+    """
+
+    with open(path, "wb") as file:
+        color = None
+
+        if image.dtype.name != "float32":
+            raise Exception("Image dtype must be float32.")
+
+        image = np.flipud(image)
+
+        if len(image.shape) == 3 and image.shape[2] == 3:  # color image
+            color = True
+        elif (
+            len(image.shape) == 2 or len(image.shape) == 3 and image.shape[2] == 1
+        ):  # greyscale
+            color = False
+        else:
+            raise Exception("Image must have H x W x 3, H x W x 1 or H x W dimensions.")
+
+        file.write("PF\n" if color else "Pf\n".encode())
+        file.write("%d %d\n".encode() % (image.shape[1], image.shape[0]))
+
+        endian = image.dtype.byteorder
+
+        if endian == "<" or endian == "=" and sys.byteorder == "little":
+            scale = -scale
+
+        file.write("%f\n".encode() % scale)
+
+        image.tofile(file)
+
+
+def read_image(path):
+    """Read image and output RGB image (0-1).
+
+    Args:
+        path (str): path to file
+
+    Returns:
+        array: RGB image (0-1)
+    """
+    img = cv2.imread(path)
+
+    if img.ndim == 2:
+        img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
+
+    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) / 255.0
+
+    return img
+
+
+def resize_image(img):
+    """Resize image and make it fit for network.
+
+    Args:
+        img (array): image
+
+    Returns:
+        tensor: data ready for network
+    """
+    height_orig = img.shape[0]
+    width_orig = img.shape[1]
+
+    if width_orig > height_orig:
+        scale = width_orig / 384
+    else:
+        scale = height_orig / 384
+
+    height = (np.ceil(height_orig / scale / 32) * 32).astype(int)
+    width = (np.ceil(width_orig / scale / 32) * 32).astype(int)
+
+    img_resized = cv2.resize(img, (width, height), interpolation=cv2.INTER_AREA)
+
+    img_resized = (
+        torch.from_numpy(np.transpose(img_resized, (2, 0, 1))).contiguous().float()
+    )
+    img_resized = img_resized.unsqueeze(0)
+
+    return img_resized
+
+
+def resize_depth(depth, width, height):
+    """Resize depth map and bring to CPU (numpy).
+
+    Args:
+        depth (tensor): depth
+        width (int): image width
+        height (int): image height
+
+    Returns:
+        array: processed depth
+    """
+    depth = torch.squeeze(depth[0, :, :, :]).to("cpu")
+
+    depth_resized = cv2.resize(
+        depth.numpy(), (width, height), interpolation=cv2.INTER_CUBIC
+    )
+
+    return depth_resized
+
+def write_depth(path, depth, bits=1):
+    """Write depth map to pfm and png file.
+
+    Args:
+        path (str): filepath without extension
+        depth (array): depth
+    """
+    write_pfm(path + ".pfm", depth.astype(np.float32))
+
+    depth_min = depth.min()
+    depth_max = depth.max()
+
+    max_val = (2**(8*bits))-1
+
+    if depth_max - depth_min > np.finfo("float").eps:
+        out = max_val * (depth - depth_min) / (depth_max - depth_min)
+    else:
+        out = np.zeros(depth.shape, dtype=depth.type)
+
+    if bits == 1:
+        cv2.imwrite(path + ".png", out.astype("uint8"))
+    elif bits == 2:
+        cv2.imwrite(path + ".png", out.astype("uint16"))
+
+    return

negi/repos/controlnet_aux/src/controlnet_aux/mlsd/LICENSE

@@ -0,0 +1,201 @@
+                                Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "{}"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2021-present NAVER Corp.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

negi/repos/controlnet_aux/src/controlnet_aux/mlsd/__init__.py

@@ -0,0 +1,79 @@
+import os
+import warnings
+
+import cv2
+import numpy as np
+import torch
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from ..util import HWC3, resize_image
+from .models.mbv2_mlsd_large import MobileV2_MLSD_Large
+from .utils import pred_lines
+
+
+class MLSDdetector:
+    def __init__(self, model):
+        self.model = model
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_or_path, filename=None, cache_dir=None):
+        if pretrained_model_or_path == "lllyasviel/ControlNet":
+            filename = filename or "annotator/ckpts/mlsd_large_512_fp32.pth"
+        else:
+            filename = filename or "mlsd_large_512_fp32.pth"
+
+        if os.path.isdir(pretrained_model_or_path):
+            model_path = os.path.join(pretrained_model_or_path, filename)
+        else:
+            model_path = hf_hub_download(pretrained_model_or_path, filename, cache_dir=cache_dir)
+
+        model = MobileV2_MLSD_Large()
+        model.load_state_dict(torch.load(model_path), strict=True)
+        model.eval()
+
+        return cls(model)
+
+    def to(self, device):
+        self.model.to(device)
+        return self
+    
+    def __call__(self, input_image, thr_v=0.1, thr_d=0.1, detect_resolution=512, image_resolution=512, output_type="pil", **kwargs):
+        if "return_pil" in kwargs:
+            warnings.warn("return_pil is deprecated. Use output_type instead.", DeprecationWarning)
+            output_type = "pil" if kwargs["return_pil"] else "np"
+        if type(output_type) is bool:
+            warnings.warn("Passing `True` or `False` to `output_type` is deprecated and will raise an error in future versions")
+            if output_type:
+                output_type = "pil"
+
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+
+        assert input_image.ndim == 3
+        img = input_image
+        img_output = np.zeros_like(img)
+        try:
+            with torch.no_grad():
+                lines = pred_lines(img, self.model, [img.shape[0], img.shape[1]], thr_v, thr_d)
+                for line in lines:
+                    x_start, y_start, x_end, y_end = [int(val) for val in line]
+                    cv2.line(img_output, (x_start, y_start), (x_end, y_end), [255, 255, 255], 1)
+        except Exception as e:
+            pass
+
+        detected_map = img_output[:, :, 0]
+        detected_map = HWC3(detected_map)
+
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+
+        return detected_map

negi/repos/controlnet_aux/src/controlnet_aux/mlsd/utils.py

@@ -0,0 +1,584 @@
+'''
+modified by  lihaoweicv
+pytorch version
+'''
+
+'''
+M-LSD
+Copyright 2021-present NAVER Corp.
+Apache License v2.0
+'''
+
+import os
+import numpy as np
+import cv2
+import torch
+from  torch.nn import functional as F
+
+
+def deccode_output_score_and_ptss(tpMap, topk_n = 200, ksize = 5):
+    '''
+    tpMap:
+    center: tpMap[1, 0, :, :]
+    displacement: tpMap[1, 1:5, :, :]
+    '''
+    b, c, h, w = tpMap.shape
+    assert  b==1, 'only support bsize==1'
+    displacement = tpMap[:, 1:5, :, :][0]
+    center = tpMap[:, 0, :, :]
+    heat = torch.sigmoid(center)
+    hmax = F.max_pool2d( heat, (ksize, ksize), stride=1, padding=(ksize-1)//2)
+    keep = (hmax == heat).float()
+    heat = heat * keep
+    heat = heat.reshape(-1, )
+
+    scores, indices = torch.topk(heat, topk_n, dim=-1, largest=True)
+    yy = torch.floor_divide(indices, w).unsqueeze(-1)
+    xx = torch.fmod(indices, w).unsqueeze(-1)
+    ptss = torch.cat((yy, xx),dim=-1)
+
+    ptss   = ptss.detach().cpu().numpy()
+    scores = scores.detach().cpu().numpy()
+    displacement = displacement.detach().cpu().numpy()
+    displacement = displacement.transpose((1,2,0))
+    return  ptss, scores, displacement
+
+
+def pred_lines(image, model,
+               input_shape=[512, 512],
+               score_thr=0.10,
+               dist_thr=20.0):
+    h, w, _ = image.shape
+
+    device = next(iter(model.parameters())).device
+    h_ratio, w_ratio = [h / input_shape[0], w / input_shape[1]]
+
+    resized_image = np.concatenate([cv2.resize(image, (input_shape[1], input_shape[0]), interpolation=cv2.INTER_AREA),
+                                    np.ones([input_shape[0], input_shape[1], 1])], axis=-1)
+
+    resized_image = resized_image.transpose((2,0,1))
+    batch_image = np.expand_dims(resized_image, axis=0).astype('float32')
+    batch_image = (batch_image / 127.5) - 1.0
+
+    batch_image = torch.from_numpy(batch_image).float()
+    batch_image = batch_image.to(device)
+    outputs = model(batch_image)
+    pts, pts_score, vmap = deccode_output_score_and_ptss(outputs, 200, 3)
+    start = vmap[:, :, :2]
+    end = vmap[:, :, 2:]
+    dist_map = np.sqrt(np.sum((start - end) ** 2, axis=-1))
+
+    segments_list = []
+    for center, score in zip(pts, pts_score):
+        y, x = center
+        distance = dist_map[y, x]
+        if score > score_thr and distance > dist_thr:
+            disp_x_start, disp_y_start, disp_x_end, disp_y_end = vmap[y, x, :]
+            x_start = x + disp_x_start
+            y_start = y + disp_y_start
+            x_end = x + disp_x_end
+            y_end = y + disp_y_end
+            segments_list.append([x_start, y_start, x_end, y_end])
+
+    lines = 2 * np.array(segments_list)  # 256 > 512
+    lines[:, 0] = lines[:, 0] * w_ratio
+    lines[:, 1] = lines[:, 1] * h_ratio
+    lines[:, 2] = lines[:, 2] * w_ratio
+    lines[:, 3] = lines[:, 3] * h_ratio
+
+    return lines
+
+
+def pred_squares(image,
+                 model,
+                 input_shape=[512, 512],
+                 params={'score': 0.06,
+                         'outside_ratio': 0.28,
+                         'inside_ratio': 0.45,
+                         'w_overlap': 0.0,
+                         'w_degree': 1.95,
+                         'w_length': 0.0,
+                         'w_area': 1.86,
+                         'w_center': 0.14}):
+    '''
+    shape = [height, width]
+    '''
+    h, w, _ = image.shape
+    original_shape = [h, w]
+    device = next(iter(model.parameters())).device
+
+    resized_image = np.concatenate([cv2.resize(image, (input_shape[0], input_shape[1]), interpolation=cv2.INTER_AREA),
+                                    np.ones([input_shape[0], input_shape[1], 1])], axis=-1)
+    resized_image = resized_image.transpose((2, 0, 1))
+    batch_image = np.expand_dims(resized_image, axis=0).astype('float32')
+    batch_image = (batch_image / 127.5) - 1.0
+
+    batch_image = torch.from_numpy(batch_image).float().to(device)
+    outputs = model(batch_image)
+
+    pts, pts_score, vmap = deccode_output_score_and_ptss(outputs, 200, 3)
+    start = vmap[:, :, :2]  # (x, y)
+    end = vmap[:, :, 2:]  # (x, y)
+    dist_map = np.sqrt(np.sum((start - end) ** 2, axis=-1))
+
+    junc_list = []
+    segments_list = []
+    for junc, score in zip(pts, pts_score):
+        y, x = junc
+        distance = dist_map[y, x]
+        if score > params['score'] and distance > 20.0:
+            junc_list.append([x, y])
+            disp_x_start, disp_y_start, disp_x_end, disp_y_end = vmap[y, x, :]
+            d_arrow = 1.0
+            x_start = x + d_arrow * disp_x_start
+            y_start = y + d_arrow * disp_y_start
+            x_end = x + d_arrow * disp_x_end
+            y_end = y + d_arrow * disp_y_end
+            segments_list.append([x_start, y_start, x_end, y_end])
+
+    segments = np.array(segments_list)
+
+    ####### post processing for squares
+    # 1. get unique lines
+    point = np.array([[0, 0]])
+    point = point[0]
+    start = segments[:, :2]
+    end = segments[:, 2:]
+    diff = start - end
+    a = diff[:, 1]
+    b = -diff[:, 0]
+    c = a * start[:, 0] + b * start[:, 1]
+
+    d = np.abs(a * point[0] + b * point[1] - c) / np.sqrt(a ** 2 + b ** 2 + 1e-10)
+    theta = np.arctan2(diff[:, 0], diff[:, 1]) * 180 / np.pi
+    theta[theta < 0.0] += 180
+    hough = np.concatenate([d[:, None], theta[:, None]], axis=-1)
+
+    d_quant = 1
+    theta_quant = 2
+    hough[:, 0] //= d_quant
+    hough[:, 1] //= theta_quant
+    _, indices, counts = np.unique(hough, axis=0, return_index=True, return_counts=True)
+
+    acc_map = np.zeros([512 // d_quant + 1, 360 // theta_quant + 1], dtype='float32')
+    idx_map = np.zeros([512 // d_quant + 1, 360 // theta_quant + 1], dtype='int32') - 1
+    yx_indices = hough[indices, :].astype('int32')
+    acc_map[yx_indices[:, 0], yx_indices[:, 1]] = counts
+    idx_map[yx_indices[:, 0], yx_indices[:, 1]] = indices
+
+    acc_map_np = acc_map
+    # acc_map = acc_map[None, :, :, None]
+    #
+    # ### fast suppression using tensorflow op
+    # acc_map = tf.constant(acc_map, dtype=tf.float32)
+    # max_acc_map = tf.keras.layers.MaxPool2D(pool_size=(5, 5), strides=1, padding='same')(acc_map)
+    # acc_map = acc_map * tf.cast(tf.math.equal(acc_map, max_acc_map), tf.float32)
+    # flatten_acc_map = tf.reshape(acc_map, [1, -1])
+    # topk_values, topk_indices = tf.math.top_k(flatten_acc_map, k=len(pts))
+    # _, h, w, _ = acc_map.shape
+    # y = tf.expand_dims(topk_indices // w, axis=-1)
+    # x = tf.expand_dims(topk_indices % w, axis=-1)
+    # yx = tf.concat([y, x], axis=-1)
+
+    ### fast suppression using pytorch op
+    acc_map = torch.from_numpy(acc_map_np).unsqueeze(0).unsqueeze(0)
+    _,_, h, w = acc_map.shape
+    max_acc_map = F.max_pool2d(acc_map,kernel_size=5, stride=1, padding=2)
+    acc_map = acc_map * ( (acc_map == max_acc_map).float() )
+    flatten_acc_map = acc_map.reshape([-1, ])
+
+    scores, indices = torch.topk(flatten_acc_map, len(pts), dim=-1, largest=True)
+    yy = torch.div(indices, w, rounding_mode='floor').unsqueeze(-1)
+    xx = torch.fmod(indices, w).unsqueeze(-1)
+    yx = torch.cat((yy, xx), dim=-1)
+
+    yx = yx.detach().cpu().numpy()
+
+    topk_values = scores.detach().cpu().numpy()
+    indices = idx_map[yx[:, 0], yx[:, 1]]
+    basis = 5 // 2
+
+    merged_segments = []
+    for yx_pt, max_indice, value in zip(yx, indices, topk_values):
+        y, x = yx_pt
+        if max_indice == -1 or value == 0:
+            continue
+        segment_list = []
+        for y_offset in range(-basis, basis + 1):
+            for x_offset in range(-basis, basis + 1):
+                indice = idx_map[y + y_offset, x + x_offset]
+                cnt = int(acc_map_np[y + y_offset, x + x_offset])
+                if indice != -1:
+                    segment_list.append(segments[indice])
+                if cnt > 1:
+                    check_cnt = 1
+                    current_hough = hough[indice]
+                    for new_indice, new_hough in enumerate(hough):
+                        if (current_hough == new_hough).all() and indice != new_indice:
+                            segment_list.append(segments[new_indice])
+                            check_cnt += 1
+                        if check_cnt == cnt:
+                            break
+        group_segments = np.array(segment_list).reshape([-1, 2])
+        sorted_group_segments = np.sort(group_segments, axis=0)
+        x_min, y_min = sorted_group_segments[0, :]
+        x_max, y_max = sorted_group_segments[-1, :]
+
+        deg = theta[max_indice]
+        if deg >= 90:
+            merged_segments.append([x_min, y_max, x_max, y_min])
+        else:
+            merged_segments.append([x_min, y_min, x_max, y_max])
+
+    # 2. get intersections
+    new_segments = np.array(merged_segments)  # (x1, y1, x2, y2)
+    start = new_segments[:, :2]  # (x1, y1)
+    end = new_segments[:, 2:]  # (x2, y2)
+    new_centers = (start + end) / 2.0
+    diff = start - end
+    dist_segments = np.sqrt(np.sum(diff ** 2, axis=-1))
+
+    # ax + by = c
+    a = diff[:, 1]
+    b = -diff[:, 0]
+    c = a * start[:, 0] + b * start[:, 1]
+    pre_det = a[:, None] * b[None, :]
+    det = pre_det - np.transpose(pre_det)
+
+    pre_inter_y = a[:, None] * c[None, :]
+    inter_y = (pre_inter_y - np.transpose(pre_inter_y)) / (det + 1e-10)
+    pre_inter_x = c[:, None] * b[None, :]
+    inter_x = (pre_inter_x - np.transpose(pre_inter_x)) / (det + 1e-10)
+    inter_pts = np.concatenate([inter_x[:, :, None], inter_y[:, :, None]], axis=-1).astype('int32')
+
+    # 3. get corner information
+    # 3.1 get distance
+    '''
+    dist_segments:
+        | dist(0), dist(1), dist(2), ...|
+    dist_inter_to_segment1:
+        | dist(inter,0), dist(inter,0), dist(inter,0), ... |
+        | dist(inter,1), dist(inter,1), dist(inter,1), ... |
+        ...
+    dist_inter_to_semgnet2:
+        | dist(inter,0), dist(inter,1), dist(inter,2), ... |
+        | dist(inter,0), dist(inter,1), dist(inter,2), ... |
+        ...
+    '''
+
+    dist_inter_to_segment1_start = np.sqrt(
+        np.sum(((inter_pts - start[:, None, :]) ** 2), axis=-1, keepdims=True))  # [n_batch, n_batch, 1]
+    dist_inter_to_segment1_end = np.sqrt(
+        np.sum(((inter_pts - end[:, None, :]) ** 2), axis=-1, keepdims=True))  # [n_batch, n_batch, 1]
+    dist_inter_to_segment2_start = np.sqrt(
+        np.sum(((inter_pts - start[None, :, :]) ** 2), axis=-1, keepdims=True))  # [n_batch, n_batch, 1]
+    dist_inter_to_segment2_end = np.sqrt(
+        np.sum(((inter_pts - end[None, :, :]) ** 2), axis=-1, keepdims=True))  # [n_batch, n_batch, 1]
+
+    # sort ascending
+    dist_inter_to_segment1 = np.sort(
+        np.concatenate([dist_inter_to_segment1_start, dist_inter_to_segment1_end], axis=-1),
+        axis=-1)  # [n_batch, n_batch, 2]
+    dist_inter_to_segment2 = np.sort(
+        np.concatenate([dist_inter_to_segment2_start, dist_inter_to_segment2_end], axis=-1),
+        axis=-1)  # [n_batch, n_batch, 2]
+
+    # 3.2 get degree
+    inter_to_start = new_centers[:, None, :] - inter_pts
+    deg_inter_to_start = np.arctan2(inter_to_start[:, :, 1], inter_to_start[:, :, 0]) * 180 / np.pi
+    deg_inter_to_start[deg_inter_to_start < 0.0] += 360
+    inter_to_end = new_centers[None, :, :] - inter_pts
+    deg_inter_to_end = np.arctan2(inter_to_end[:, :, 1], inter_to_end[:, :, 0]) * 180 / np.pi
+    deg_inter_to_end[deg_inter_to_end < 0.0] += 360
+
+    '''
+    B -- G
+    |    |
+    C -- R
+    B : blue / G: green / C: cyan / R: red
+
+    0 -- 1
+    |    |
+    3 -- 2
+    '''
+    # rename variables
+    deg1_map, deg2_map = deg_inter_to_start, deg_inter_to_end
+    # sort deg ascending
+    deg_sort = np.sort(np.concatenate([deg1_map[:, :, None], deg2_map[:, :, None]], axis=-1), axis=-1)
+
+    deg_diff_map = np.abs(deg1_map - deg2_map)
+    # we only consider the smallest degree of intersect
+    deg_diff_map[deg_diff_map > 180] = 360 - deg_diff_map[deg_diff_map > 180]
+
+    # define available degree range
+    deg_range = [60, 120]
+
+    corner_dict = {corner_info: [] for corner_info in range(4)}
+    inter_points = []
+    for i in range(inter_pts.shape[0]):
+        for j in range(i + 1, inter_pts.shape[1]):
+            # i, j > line index, always i < j
+            x, y = inter_pts[i, j, :]
+            deg1, deg2 = deg_sort[i, j, :]
+            deg_diff = deg_diff_map[i, j]
+
+            check_degree = deg_diff > deg_range[0] and deg_diff < deg_range[1]
+
+            outside_ratio = params['outside_ratio']  # over ratio >>> drop it!
+            inside_ratio = params['inside_ratio']  # over ratio >>> drop it!
+            check_distance = ((dist_inter_to_segment1[i, j, 1] >= dist_segments[i] and \
+                               dist_inter_to_segment1[i, j, 0] <= dist_segments[i] * outside_ratio) or \
+                              (dist_inter_to_segment1[i, j, 1] <= dist_segments[i] and \
+                               dist_inter_to_segment1[i, j, 0] <= dist_segments[i] * inside_ratio)) and \
+                             ((dist_inter_to_segment2[i, j, 1] >= dist_segments[j] and \
+                               dist_inter_to_segment2[i, j, 0] <= dist_segments[j] * outside_ratio) or \
+                              (dist_inter_to_segment2[i, j, 1] <= dist_segments[j] and \
+                               dist_inter_to_segment2[i, j, 0] <= dist_segments[j] * inside_ratio))
+
+            if check_degree and check_distance:
+                corner_info = None
+
+                if (deg1 >= 0 and deg1 <= 45 and deg2 >= 45 and deg2 <= 120) or \
+                        (deg2 >= 315 and deg1 >= 45 and deg1 <= 120):
+                    corner_info, color_info = 0, 'blue'
+                elif (deg1 >= 45 and deg1 <= 125 and deg2 >= 125 and deg2 <= 225):
+                    corner_info, color_info = 1, 'green'
+                elif (deg1 >= 125 and deg1 <= 225 and deg2 >= 225 and deg2 <= 315):
+                    corner_info, color_info = 2, 'black'
+                elif (deg1 >= 0 and deg1 <= 45 and deg2 >= 225 and deg2 <= 315) or \
+                        (deg2 >= 315 and deg1 >= 225 and deg1 <= 315):
+                    corner_info, color_info = 3, 'cyan'
+                else:
+                    corner_info, color_info = 4, 'red'  # we don't use it
+                    continue
+
+                corner_dict[corner_info].append([x, y, i, j])
+                inter_points.append([x, y])
+
+    square_list = []
+    connect_list = []
+    segments_list = []
+    for corner0 in corner_dict[0]:
+        for corner1 in corner_dict[1]:
+            connect01 = False
+            for corner0_line in corner0[2:]:
+                if corner0_line in corner1[2:]:
+                    connect01 = True
+                    break
+            if connect01:
+                for corner2 in corner_dict[2]:
+                    connect12 = False
+                    for corner1_line in corner1[2:]:
+                        if corner1_line in corner2[2:]:
+                            connect12 = True
+                            break
+                    if connect12:
+                        for corner3 in corner_dict[3]:
+                            connect23 = False
+                            for corner2_line in corner2[2:]:
+                                if corner2_line in corner3[2:]:
+                                    connect23 = True
+                                    break
+                            if connect23:
+                                for corner3_line in corner3[2:]:
+                                    if corner3_line in corner0[2:]:
+                                        # SQUARE!!!
+                                        '''
+                                        0 -- 1
+                                        |    |
+                                        3 -- 2
+                                        square_list:
+                                            order: 0 > 1 > 2 > 3
+                                            | x0, y0, x1, y1, x2, y2, x3, y3 |
+                                            | x0, y0, x1, y1, x2, y2, x3, y3 |
+                                            ...
+                                        connect_list:
+                                            order: 01 > 12 > 23 > 30
+                                            | line_idx01, line_idx12, line_idx23, line_idx30 |
+                                            | line_idx01, line_idx12, line_idx23, line_idx30 |
+                                            ...
+                                        segments_list:
+                                            order: 0 > 1 > 2 > 3
+                                            | line_idx0_i, line_idx0_j, line_idx1_i, line_idx1_j, line_idx2_i, line_idx2_j, line_idx3_i, line_idx3_j |
+                                            | line_idx0_i, line_idx0_j, line_idx1_i, line_idx1_j, line_idx2_i, line_idx2_j, line_idx3_i, line_idx3_j |
+                                            ...
+                                        '''
+                                        square_list.append(corner0[:2] + corner1[:2] + corner2[:2] + corner3[:2])
+                                        connect_list.append([corner0_line, corner1_line, corner2_line, corner3_line])
+                                        segments_list.append(corner0[2:] + corner1[2:] + corner2[2:] + corner3[2:])
+
+    def check_outside_inside(segments_info, connect_idx):
+        # return 'outside or inside', min distance, cover_param, peri_param
+        if connect_idx == segments_info[0]:
+            check_dist_mat = dist_inter_to_segment1
+        else:
+            check_dist_mat = dist_inter_to_segment2
+
+        i, j = segments_info
+        min_dist, max_dist = check_dist_mat[i, j, :]
+        connect_dist = dist_segments[connect_idx]
+        if max_dist > connect_dist:
+            return 'outside', min_dist, 0, 1
+        else:
+            return 'inside', min_dist, -1, -1
+
+    top_square = None
+
+    try:
+        map_size = input_shape[0] / 2
+        squares = np.array(square_list).reshape([-1, 4, 2])
+        score_array = []
+        connect_array = np.array(connect_list)
+        segments_array = np.array(segments_list).reshape([-1, 4, 2])
+
+        # get degree of corners:
+        squares_rollup = np.roll(squares, 1, axis=1)
+        squares_rolldown = np.roll(squares, -1, axis=1)
+        vec1 = squares_rollup - squares
+        normalized_vec1 = vec1 / (np.linalg.norm(vec1, axis=-1, keepdims=True) + 1e-10)
+        vec2 = squares_rolldown - squares
+        normalized_vec2 = vec2 / (np.linalg.norm(vec2, axis=-1, keepdims=True) + 1e-10)
+        inner_products = np.sum(normalized_vec1 * normalized_vec2, axis=-1)  # [n_squares, 4]
+        squares_degree = np.arccos(inner_products) * 180 / np.pi  # [n_squares, 4]
+
+        # get square score
+        overlap_scores = []
+        degree_scores = []
+        length_scores = []
+
+        for connects, segments, square, degree in zip(connect_array, segments_array, squares, squares_degree):
+            '''
+            0 -- 1
+            |    |
+            3 -- 2
+
+            # segments: [4, 2]
+            # connects: [4]
+            '''
+
+            ###################################### OVERLAP SCORES
+            cover = 0
+            perimeter = 0
+            # check 0 > 1 > 2 > 3
+            square_length = []
+
+            for start_idx in range(4):
+                end_idx = (start_idx + 1) % 4
+
+                connect_idx = connects[start_idx]  # segment idx of segment01
+                start_segments = segments[start_idx]
+                end_segments = segments[end_idx]
+
+                start_point = square[start_idx]
+                end_point = square[end_idx]
+
+                # check whether outside or inside
+                start_position, start_min, start_cover_param, start_peri_param = check_outside_inside(start_segments,
+                                                                                                      connect_idx)
+                end_position, end_min, end_cover_param, end_peri_param = check_outside_inside(end_segments, connect_idx)
+
+                cover += dist_segments[connect_idx] + start_cover_param * start_min + end_cover_param * end_min
+                perimeter += dist_segments[connect_idx] + start_peri_param * start_min + end_peri_param * end_min
+
+                square_length.append(
+                    dist_segments[connect_idx] + start_peri_param * start_min + end_peri_param * end_min)
+
+            overlap_scores.append(cover / perimeter)
+            ######################################
+            ###################################### DEGREE SCORES
+            '''
+            deg0 vs deg2
+            deg1 vs deg3
+            '''
+            deg0, deg1, deg2, deg3 = degree
+            deg_ratio1 = deg0 / deg2
+            if deg_ratio1 > 1.0:
+                deg_ratio1 = 1 / deg_ratio1
+            deg_ratio2 = deg1 / deg3
+            if deg_ratio2 > 1.0:
+                deg_ratio2 = 1 / deg_ratio2
+            degree_scores.append((deg_ratio1 + deg_ratio2) / 2)
+            ######################################
+            ###################################### LENGTH SCORES
+            '''
+            len0 vs len2
+            len1 vs len3
+            '''
+            len0, len1, len2, len3 = square_length
+            len_ratio1 = len0 / len2 if len2 > len0 else len2 / len0
+            len_ratio2 = len1 / len3 if len3 > len1 else len3 / len1
+            length_scores.append((len_ratio1 + len_ratio2) / 2)
+
+            ######################################
+
+        overlap_scores = np.array(overlap_scores)
+        overlap_scores /= np.max(overlap_scores)
+
+        degree_scores = np.array(degree_scores)
+        # degree_scores /= np.max(degree_scores)
+
+        length_scores = np.array(length_scores)
+
+        ###################################### AREA SCORES
+        area_scores = np.reshape(squares, [-1, 4, 2])
+        area_x = area_scores[:, :, 0]
+        area_y = area_scores[:, :, 1]
+        correction = area_x[:, -1] * area_y[:, 0] - area_y[:, -1] * area_x[:, 0]
+        area_scores = np.sum(area_x[:, :-1] * area_y[:, 1:], axis=-1) - np.sum(area_y[:, :-1] * area_x[:, 1:], axis=-1)
+        area_scores = 0.5 * np.abs(area_scores + correction)
+        area_scores /= (map_size * map_size)  # np.max(area_scores)
+        ######################################
+
+        ###################################### CENTER SCORES
+        centers = np.array([[256 // 2, 256 // 2]], dtype='float32')  # [1, 2]
+        # squares: [n, 4, 2]
+        square_centers = np.mean(squares, axis=1)  # [n, 2]
+        center2center = np.sqrt(np.sum((centers - square_centers) ** 2))
+        center_scores = center2center / (map_size / np.sqrt(2.0))
+
+        '''
+        score_w = [overlap, degree, area, center, length]
+        '''
+        score_w = [0.0, 1.0, 10.0, 0.5, 1.0]
+        score_array = params['w_overlap'] * overlap_scores \
+                      + params['w_degree'] * degree_scores \
+                      + params['w_area'] * area_scores \
+                      - params['w_center'] * center_scores \
+                      + params['w_length'] * length_scores
+
+        best_square = []
+
+        sorted_idx = np.argsort(score_array)[::-1]
+        score_array = score_array[sorted_idx]
+        squares = squares[sorted_idx]
+
+    except Exception as e:
+        pass
+
+    '''return list
+    merged_lines, squares, scores
+    '''
+
+    try:
+        new_segments[:, 0] = new_segments[:, 0] * 2 / input_shape[1] * original_shape[1]
+        new_segments[:, 1] = new_segments[:, 1] * 2 / input_shape[0] * original_shape[0]
+        new_segments[:, 2] = new_segments[:, 2] * 2 / input_shape[1] * original_shape[1]
+        new_segments[:, 3] = new_segments[:, 3] * 2 / input_shape[0] * original_shape[0]
+    except:
+        new_segments = []
+
+    try:
+        squares[:, :, 0] = squares[:, :, 0] * 2 / input_shape[1] * original_shape[1]
+        squares[:, :, 1] = squares[:, :, 1] * 2 / input_shape[0] * original_shape[0]
+    except:
+        squares = []
+        score_array = []
+
+    try:
+        inter_points = np.array(inter_points)
+        inter_points[:, 0] = inter_points[:, 0] * 2 / input_shape[1] * original_shape[1]
+        inter_points[:, 1] = inter_points[:, 1] * 2 / input_shape[0] * original_shape[0]
+    except:
+        inter_points = []
+
+    return new_segments, squares, score_array, inter_points

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2022 Caroline Chan
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/__init__.py

@@ -0,0 +1,109 @@
+import os
+import types
+import warnings
+
+import cv2
+import numpy as np
+import torch
+import torchvision.transforms as transforms
+from einops import rearrange
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from ..util import HWC3, resize_image
+from .nets.NNET import NNET
+
+
+# load model
+def load_checkpoint(fpath, model):
+    ckpt = torch.load(fpath, map_location='cpu')['model']
+
+    load_dict = {}
+    for k, v in ckpt.items():
+        if k.startswith('module.'):
+            k_ = k.replace('module.', '')
+            load_dict[k_] = v
+        else:
+            load_dict[k] = v
+
+    model.load_state_dict(load_dict)
+    return model
+
+class NormalBaeDetector:
+    def __init__(self, model):
+        self.model = model
+        self.norm = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_or_path, filename=None, cache_dir=None):
+        filename = filename or "scannet.pt"
+
+        if os.path.isdir(pretrained_model_or_path):
+            model_path = os.path.join(pretrained_model_or_path, filename)
+        else:
+            model_path = hf_hub_download(pretrained_model_or_path, filename, cache_dir=cache_dir)
+
+        args = types.SimpleNamespace()
+        args.mode = 'client'
+        args.architecture = 'BN'
+        args.pretrained = 'scannet'
+        args.sampling_ratio = 0.4
+        args.importance_ratio = 0.7
+        model = NNET(args)
+        model = load_checkpoint(model_path, model)
+        model.eval()
+
+        return cls(model)
+
+    def to(self, device):
+        self.model.to(device)
+        return self
+
+
+    def __call__(self, input_image, detect_resolution=512, image_resolution=512, output_type="pil", **kwargs):
+        if "return_pil" in kwargs:
+            warnings.warn("return_pil is deprecated. Use output_type instead.", DeprecationWarning)
+            output_type = "pil" if kwargs["return_pil"] else "np"
+        if type(output_type) is bool:
+            warnings.warn("Passing `True` or `False` to `output_type` is deprecated and will raise an error in future versions")
+            if output_type:
+                output_type = "pil"
+
+        device = next(iter(self.model.parameters())).device
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+
+        assert input_image.ndim == 3
+        image_normal = input_image
+        with torch.no_grad():
+            image_normal = torch.from_numpy(image_normal).float().to(device)
+            image_normal = image_normal / 255.0
+            image_normal = rearrange(image_normal, 'h w c -> 1 c h w')
+            image_normal = self.norm(image_normal)
+
+            normal = self.model(image_normal)
+            normal = normal[0][-1][:, :3]
+            # d = torch.sum(normal ** 2.0, dim=1, keepdim=True) ** 0.5
+            # d = torch.maximum(d, torch.ones_like(d) * 1e-5)
+            # normal /= d
+            normal = ((normal + 1) * 0.5).clip(0, 1)
+
+            normal = rearrange(normal[0], 'c h w -> h w c').cpu().numpy()
+            normal_image = (normal * 255.0).clip(0, 255).astype(np.uint8)
+
+        detected_map = normal_image
+        detected_map = HWC3(detected_map)      
+
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+
+        return detected_map
+    

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/NNET.py

@@ -0,0 +1,22 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .submodules.encoder import Encoder
+from .submodules.decoder import Decoder
+
+
+class NNET(nn.Module):
+    def __init__(self, args):
+        super(NNET, self).__init__()
+        self.encoder = Encoder()
+        self.decoder = Decoder(args)
+
+    def get_1x_lr_params(self):  # lr/10 learning rate
+        return self.encoder.parameters()
+
+    def get_10x_lr_params(self):  # lr learning rate
+        return self.decoder.parameters()
+
+    def forward(self, img, **kwargs):
+        return self.decoder(self.encoder(img), **kwargs)

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/baseline.py

@@ -0,0 +1,85 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .submodules.submodules import UpSampleBN, norm_normalize
+
+
+# This is the baseline encoder-decoder we used in the ablation study
+class NNET(nn.Module):
+    def __init__(self, args=None):
+        super(NNET, self).__init__()
+        self.encoder = Encoder()
+        self.decoder = Decoder(num_classes=4)
+
+    def forward(self, x, **kwargs):
+        out = self.decoder(self.encoder(x), **kwargs)
+
+        # Bilinearly upsample the output to match the input resolution
+        up_out = F.interpolate(out, size=[x.size(2), x.size(3)], mode='bilinear', align_corners=False)
+        
+        # L2-normalize the first three channels / ensure positive value for concentration parameters (kappa)
+        up_out = norm_normalize(up_out) 
+        return up_out
+
+    def get_1x_lr_params(self):  # lr/10 learning rate
+        return self.encoder.parameters()
+
+    def get_10x_lr_params(self):  # lr learning rate
+        modules = [self.decoder]
+        for m in modules:
+            yield from m.parameters()
+
+
+# Encoder
+class Encoder(nn.Module):
+    def __init__(self):
+        super(Encoder, self).__init__()
+
+        basemodel_name = 'tf_efficientnet_b5_ap'
+        basemodel = torch.hub.load('rwightman/gen-efficientnet-pytorch', basemodel_name, pretrained=True)
+
+        # Remove last layer
+        basemodel.global_pool = nn.Identity()
+        basemodel.classifier = nn.Identity()
+
+        self.original_model = basemodel
+
+    def forward(self, x):
+        features = [x]
+        for k, v in self.original_model._modules.items():
+            if (k == 'blocks'):
+                for ki, vi in v._modules.items():
+                    features.append(vi(features[-1]))
+            else:
+                features.append(v(features[-1]))
+        return features
+
+
+# Decoder (no pixel-wise MLP, no uncertainty-guided sampling)
+class Decoder(nn.Module):
+    def __init__(self, num_classes=4):
+        super(Decoder, self).__init__()
+        self.conv2 = nn.Conv2d(2048, 2048, kernel_size=1, stride=1, padding=0)
+        self.up1 = UpSampleBN(skip_input=2048 + 176, output_features=1024)
+        self.up2 = UpSampleBN(skip_input=1024 + 64, output_features=512)
+        self.up3 = UpSampleBN(skip_input=512 + 40, output_features=256)
+        self.up4 = UpSampleBN(skip_input=256 + 24, output_features=128)
+        self.conv3 = nn.Conv2d(128, num_classes, kernel_size=3, stride=1, padding=1)
+
+    def forward(self, features):
+        x_block0, x_block1, x_block2, x_block3, x_block4 = features[4], features[5], features[6], features[8], features[11]
+        x_d0 = self.conv2(x_block4)
+        x_d1 = self.up1(x_d0, x_block3)
+        x_d2 = self.up2(x_d1, x_block2)
+        x_d3 = self.up3(x_d2, x_block1)
+        x_d4 = self.up4(x_d3, x_block0)
+        out = self.conv3(x_d4)
+        return out
+
+
+if __name__ == '__main__':
+    model = Baseline()
+    x = torch.rand(2, 3, 480, 640)
+    out = model(x)
+    print(out.shape)

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/decoder.py

@@ -0,0 +1,202 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from .submodules import UpSampleBN, UpSampleGN, norm_normalize, sample_points
+
+
+class Decoder(nn.Module):
+    def __init__(self, args):
+        super(Decoder, self).__init__()
+
+        # hyper-parameter for sampling
+        self.sampling_ratio = args.sampling_ratio
+        self.importance_ratio = args.importance_ratio
+
+        # feature-map
+        self.conv2 = nn.Conv2d(2048, 2048, kernel_size=1, stride=1, padding=0)
+        if args.architecture == 'BN':
+            self.up1 = UpSampleBN(skip_input=2048 + 176, output_features=1024)
+            self.up2 = UpSampleBN(skip_input=1024 + 64, output_features=512)
+            self.up3 = UpSampleBN(skip_input=512 + 40, output_features=256)
+            self.up4 = UpSampleBN(skip_input=256 + 24, output_features=128)
+
+        elif args.architecture == 'GN':
+            self.up1 = UpSampleGN(skip_input=2048 + 176, output_features=1024)
+            self.up2 = UpSampleGN(skip_input=1024 + 64, output_features=512)
+            self.up3 = UpSampleGN(skip_input=512 + 40, output_features=256)
+            self.up4 = UpSampleGN(skip_input=256 + 24, output_features=128)
+
+        else:
+            raise Exception('invalid architecture')
+
+        # produces 1/8 res output
+        self.out_conv_res8 = nn.Conv2d(512, 4, kernel_size=3, stride=1, padding=1)
+
+        # produces 1/4 res output
+        self.out_conv_res4 = nn.Sequential(
+            nn.Conv1d(512 + 4, 128, kernel_size=1), nn.ReLU(),
+            nn.Conv1d(128, 128, kernel_size=1), nn.ReLU(),
+            nn.Conv1d(128, 128, kernel_size=1), nn.ReLU(),
+            nn.Conv1d(128, 4, kernel_size=1),
+        )
+
+        # produces 1/2 res output
+        self.out_conv_res2 = nn.Sequential(
+            nn.Conv1d(256 + 4, 128, kernel_size=1), nn.ReLU(),
+            nn.Conv1d(128, 128, kernel_size=1), nn.ReLU(),
+            nn.Conv1d(128, 128, kernel_size=1), nn.ReLU(),
+            nn.Conv1d(128, 4, kernel_size=1),
+        )
+
+        # produces 1/1 res output
+        self.out_conv_res1 = nn.Sequential(
+            nn.Conv1d(128 + 4, 128, kernel_size=1), nn.ReLU(),
+            nn.Conv1d(128, 128, kernel_size=1), nn.ReLU(),
+            nn.Conv1d(128, 128, kernel_size=1), nn.ReLU(),
+            nn.Conv1d(128, 4, kernel_size=1),
+        )
+
+    def forward(self, features, gt_norm_mask=None, mode='test'):
+        x_block0, x_block1, x_block2, x_block3, x_block4 = features[4], features[5], features[6], features[8], features[11]
+
+        # generate feature-map
+
+        x_d0 = self.conv2(x_block4)                     # x_d0 : [2, 2048, 15, 20]      1/32 res
+        x_d1 = self.up1(x_d0, x_block3)                 # x_d1 : [2, 1024, 30, 40]      1/16 res
+        x_d2 = self.up2(x_d1, x_block2)                 # x_d2 : [2, 512, 60, 80]       1/8 res
+        x_d3 = self.up3(x_d2, x_block1)                 # x_d3: [2, 256, 120, 160]      1/4 res
+        x_d4 = self.up4(x_d3, x_block0)                 # x_d4: [2, 128, 240, 320]      1/2 res
+
+        # 1/8 res output
+        out_res8 = self.out_conv_res8(x_d2)             # out_res8: [2, 4, 60, 80]      1/8 res output
+        out_res8 = norm_normalize(out_res8)             # out_res8: [2, 4, 60, 80]      1/8 res output
+
+        ################################################################################################################
+        # out_res4
+        ################################################################################################################
+
+        if mode == 'train':
+            # upsampling ... out_res8: [2, 4, 60, 80] -> out_res8_res4: [2, 4, 120, 160]
+            out_res8_res4 = F.interpolate(out_res8, scale_factor=2, mode='bilinear', align_corners=True)
+            B, _, H, W = out_res8_res4.shape
+
+            # samples: [B, 1, N, 2]
+            point_coords_res4, rows_int, cols_int = sample_points(out_res8_res4.detach(), gt_norm_mask,
+                                                                  sampling_ratio=self.sampling_ratio,
+                                                                  beta=self.importance_ratio)
+
+            # output (needed for evaluation / visualization)
+            out_res4 = out_res8_res4
+
+            # grid_sample feature-map
+            feat_res4 = F.grid_sample(x_d2, point_coords_res4, mode='bilinear', align_corners=True)  # (B, 512, 1, N)
+            init_pred = F.grid_sample(out_res8, point_coords_res4, mode='bilinear', align_corners=True)  # (B, 4, 1, N)
+            feat_res4 = torch.cat([feat_res4, init_pred], dim=1)  # (B, 512+4, 1, N)
+
+            # prediction (needed to compute loss)
+            samples_pred_res4 = self.out_conv_res4(feat_res4[:, :, 0, :])  # (B, 4, N)
+            samples_pred_res4 = norm_normalize(samples_pred_res4)  # (B, 4, N) - normalized
+
+            for i in range(B):
+                out_res4[i, :, rows_int[i, :], cols_int[i, :]] = samples_pred_res4[i, :, :]
+
+        else:
+            # grid_sample feature-map
+            feat_map = F.interpolate(x_d2, scale_factor=2, mode='bilinear', align_corners=True)
+            init_pred = F.interpolate(out_res8, scale_factor=2, mode='bilinear', align_corners=True)
+            feat_map = torch.cat([feat_map, init_pred], dim=1)  # (B, 512+4, H, W)
+            B, _, H, W = feat_map.shape
+
+            # try all pixels
+            out_res4 = self.out_conv_res4(feat_map.view(B, 512 + 4, -1))  # (B, 4, N)
+            out_res4 = norm_normalize(out_res4)  # (B, 4, N) - normalized
+            out_res4 = out_res4.view(B, 4, H, W)
+            samples_pred_res4 = point_coords_res4 = None
+
+        ################################################################################################################
+        # out_res2
+        ################################################################################################################
+
+        if mode == 'train':
+
+            # upsampling ... out_res4: [2, 4, 120, 160] -> out_res4_res2: [2, 4, 240, 320]
+            out_res4_res2 = F.interpolate(out_res4, scale_factor=2, mode='bilinear', align_corners=True)
+            B, _, H, W = out_res4_res2.shape
+
+            # samples: [B, 1, N, 2]
+            point_coords_res2, rows_int, cols_int = sample_points(out_res4_res2.detach(), gt_norm_mask,
+                                                                  sampling_ratio=self.sampling_ratio,
+                                                                  beta=self.importance_ratio)
+
+            # output (needed for evaluation / visualization)
+            out_res2 = out_res4_res2
+
+            # grid_sample feature-map
+            feat_res2 = F.grid_sample(x_d3, point_coords_res2, mode='bilinear', align_corners=True)  # (B, 256, 1, N)
+            init_pred = F.grid_sample(out_res4, point_coords_res2, mode='bilinear', align_corners=True)  # (B, 4, 1, N)
+            feat_res2 = torch.cat([feat_res2, init_pred], dim=1)  # (B, 256+4, 1, N)
+
+            # prediction (needed to compute loss)
+            samples_pred_res2 = self.out_conv_res2(feat_res2[:, :, 0, :])  # (B, 4, N)
+            samples_pred_res2 = norm_normalize(samples_pred_res2)  # (B, 4, N) - normalized
+
+            for i in range(B):
+                out_res2[i, :, rows_int[i, :], cols_int[i, :]] = samples_pred_res2[i, :, :]
+
+        else:
+            # grid_sample feature-map
+            feat_map = F.interpolate(x_d3, scale_factor=2, mode='bilinear', align_corners=True)
+            init_pred = F.interpolate(out_res4, scale_factor=2, mode='bilinear', align_corners=True)
+            feat_map = torch.cat([feat_map, init_pred], dim=1)  # (B, 512+4, H, W)
+            B, _, H, W = feat_map.shape
+
+            out_res2 = self.out_conv_res2(feat_map.view(B, 256 + 4, -1))  # (B, 4, N)
+            out_res2 = norm_normalize(out_res2)  # (B, 4, N) - normalized
+            out_res2 = out_res2.view(B, 4, H, W)
+            samples_pred_res2 = point_coords_res2 = None
+
+        ################################################################################################################
+        # out_res1
+        ################################################################################################################
+
+        if mode == 'train':
+            # upsampling ... out_res4: [2, 4, 120, 160] -> out_res4_res2: [2, 4, 240, 320]
+            out_res2_res1 = F.interpolate(out_res2, scale_factor=2, mode='bilinear', align_corners=True)
+            B, _, H, W = out_res2_res1.shape
+
+            # samples: [B, 1, N, 2]
+            point_coords_res1, rows_int, cols_int = sample_points(out_res2_res1.detach(), gt_norm_mask,
+                                                                  sampling_ratio=self.sampling_ratio,
+                                                                  beta=self.importance_ratio)
+
+            # output (needed for evaluation / visualization)
+            out_res1 = out_res2_res1
+
+            # grid_sample feature-map
+            feat_res1 = F.grid_sample(x_d4, point_coords_res1, mode='bilinear', align_corners=True)  # (B, 128, 1, N)
+            init_pred = F.grid_sample(out_res2, point_coords_res1, mode='bilinear', align_corners=True)  # (B, 4, 1, N)
+            feat_res1 = torch.cat([feat_res1, init_pred], dim=1)  # (B, 128+4, 1, N)
+
+            # prediction (needed to compute loss)
+            samples_pred_res1 = self.out_conv_res1(feat_res1[:, :, 0, :])  # (B, 4, N)
+            samples_pred_res1 = norm_normalize(samples_pred_res1)  # (B, 4, N) - normalized
+
+            for i in range(B):
+                out_res1[i, :, rows_int[i, :], cols_int[i, :]] = samples_pred_res1[i, :, :]
+
+        else:
+            # grid_sample feature-map
+            feat_map = F.interpolate(x_d4, scale_factor=2, mode='bilinear', align_corners=True)
+            init_pred = F.interpolate(out_res2, scale_factor=2, mode='bilinear', align_corners=True)
+            feat_map = torch.cat([feat_map, init_pred], dim=1)  # (B, 512+4, H, W)
+            B, _, H, W = feat_map.shape
+
+            out_res1 = self.out_conv_res1(feat_map.view(B, 128 + 4, -1))  # (B, 4, N)
+            out_res1 = norm_normalize(out_res1)  # (B, 4, N) - normalized
+            out_res1 = out_res1.view(B, 4, H, W)
+            samples_pred_res1 = point_coords_res1 = None
+
+        return [out_res8, out_res4, out_res2, out_res1], \
+               [out_res8, samples_pred_res4, samples_pred_res2, samples_pred_res1], \
+               [None, point_coords_res4, point_coords_res2, point_coords_res1]
+

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/.gitignore

@@ -0,0 +1,109 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# pyenv
+.python-version
+
+# celery beat schedule file
+celerybeat-schedule
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# pytorch stuff
+*.pth
+*.onnx
+*.pb
+
+trained_models/
+.fuse_hidden*

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/BENCHMARK.md

@@ -0,0 +1,555 @@
+# Model Performance Benchmarks
+
+All benchmarks run as per:
+
+```
+python onnx_export.py --model mobilenetv3_100 ./mobilenetv3_100.onnx
+python onnx_optimize.py ./mobilenetv3_100.onnx --output mobilenetv3_100-opt.onnx
+python onnx_to_caffe.py ./mobilenetv3_100.onnx --c2-prefix mobilenetv3
+python onnx_to_caffe.py ./mobilenetv3_100-opt.onnx --c2-prefix mobilenetv3-opt
+python caffe2_benchmark.py --c2-init ./mobilenetv3.init.pb --c2-predict ./mobilenetv3.predict.pb
+python caffe2_benchmark.py --c2-init ./mobilenetv3-opt.init.pb --c2-predict ./mobilenetv3-opt.predict.pb
+```
+
+## EfficientNet-B0
+
+### Unoptimized
+```
+Main run finished. Milliseconds per iter: 49.2862. Iters per second: 20.2897
+Time per operator type:
+        29.7378 ms.    60.5145%. Conv
+        12.1785 ms.    24.7824%. Sigmoid
+        3.62811 ms.    7.38297%. SpatialBN
+        2.98444 ms.    6.07314%. Mul
+       0.326902 ms.   0.665225%. AveragePool
+       0.197317 ms.   0.401528%. FC
+      0.0852877 ms.   0.173555%. Add
+      0.0032607 ms. 0.00663532%. Squeeze
+        49.1416 ms in Total
+FLOP per operator type:
+        0.76907 GFLOP.    95.2696%. Conv
+      0.0269508 GFLOP.    3.33857%. SpatialBN
+     0.00846444 GFLOP.    1.04855%. Mul
+       0.002561 GFLOP.   0.317248%. FC
+    0.000210112 GFLOP.  0.0260279%. Add
+       0.807256 GFLOP in Total
+Feature Memory Read per operator type:
+        58.5253 MB.    43.0891%. Mul
+        43.2015 MB.     31.807%. Conv
+        27.2869 MB.    20.0899%. SpatialBN
+        5.12912 MB.    3.77631%. FC
+         1.6809 MB.    1.23756%. Add
+        135.824 MB in Total
+Feature Memory Written per operator type:
+        33.8578 MB.    38.1965%. Mul
+        26.9881 MB.    30.4465%. Conv
+        26.9508 MB.    30.4044%. SpatialBN
+       0.840448 MB.   0.948147%. Add
+          0.004 MB. 0.00451258%. FC
+        88.6412 MB in Total
+Parameter Memory per operator type:
+        15.8248 MB.    74.9391%. Conv
+          5.124 MB.     24.265%. FC
+       0.168064 MB.   0.795877%. SpatialBN
+              0 MB.          0%. Add
+              0 MB.          0%. Mul
+        21.1168 MB in Total
+```
+### Optimized
+```
+Main run finished. Milliseconds per iter: 46.0838. Iters per second: 21.6996
+Time per operator type:
+         29.776 ms.     65.002%. Conv
+        12.2803 ms.    26.8084%. Sigmoid
+        3.15073 ms.    6.87815%. Mul
+       0.328651 ms.   0.717456%. AveragePool
+       0.186237 ms.   0.406563%. FC
+      0.0832429 ms.   0.181722%. Add
+      0.0026184 ms. 0.00571606%. Squeeze
+        45.8078 ms in Total
+FLOP per operator type:
+        0.76907 GFLOP.    98.5601%. Conv
+     0.00846444 GFLOP.    1.08476%. Mul
+       0.002561 GFLOP.   0.328205%. FC
+    0.000210112 GFLOP.  0.0269269%. Add
+       0.780305 GFLOP in Total
+Feature Memory Read per operator type:
+        58.5253 MB.    53.8803%. Mul
+        43.2855 MB.    39.8501%. Conv
+        5.12912 MB.    4.72204%. FC
+         1.6809 MB.    1.54749%. Add
+        108.621 MB in Total
+Feature Memory Written per operator type:
+        33.8578 MB.    54.8834%. Mul
+        26.9881 MB.    43.7477%. Conv
+       0.840448 MB.    1.36237%. Add
+          0.004 MB. 0.00648399%. FC
+        61.6904 MB in Total
+Parameter Memory per operator type:
+        15.8248 MB.    75.5403%. Conv
+          5.124 MB.    24.4597%. FC
+              0 MB.          0%. Add
+              0 MB.          0%. Mul
+        20.9488 MB in Total
+```
+
+## EfficientNet-B1
+### Optimized
+```
+Main run finished. Milliseconds per iter: 71.8102. Iters per second: 13.9256
+Time per operator type:
+        45.7915 ms.    66.3206%. Conv
+        17.8718 ms.    25.8841%. Sigmoid
+        4.44132 ms.    6.43244%. Mul
+        0.51001 ms.   0.738658%. AveragePool
+       0.233283 ms.   0.337868%. Add
+       0.194986 ms.   0.282402%. FC
+     0.00268255 ms. 0.00388519%. Squeeze
+        69.0456 ms in Total
+FLOP per operator type:
+        1.37105 GFLOP.    98.7673%. Conv
+      0.0138759 GFLOP.    0.99959%. Mul
+       0.002561 GFLOP.   0.184489%. FC
+    0.000674432 GFLOP.  0.0485847%. Add
+        1.38816 GFLOP in Total
+Feature Memory Read per operator type:
+         94.624 MB.    54.0789%. Mul
+        69.8255 MB.    39.9062%. Conv
+        5.39546 MB.    3.08357%. Add
+        5.12912 MB.    2.93136%. FC
+        174.974 MB in Total
+Feature Memory Written per operator type:
+        55.5035 MB.     54.555%. Mul
+        43.5333 MB.    42.7894%. Conv
+        2.69773 MB.    2.65163%. Add
+          0.004 MB. 0.00393165%. FC
+        101.739 MB in Total
+Parameter Memory per operator type:
+        25.7479 MB.    83.4024%. Conv
+          5.124 MB.    16.5976%. FC
+              0 MB.          0%. Add
+              0 MB.          0%. Mul
+        30.8719 MB in Total
+```
+
+## EfficientNet-B2
+### Optimized
+```
+Main run finished. Milliseconds per iter: 92.28. Iters per second: 10.8366
+Time per operator type:
+        61.4627 ms.    67.5845%. Conv
+        22.7458 ms.    25.0113%. Sigmoid
+        5.59931 ms.    6.15701%. Mul
+       0.642567 ms.   0.706568%. AveragePool
+       0.272795 ms.   0.299965%. Add
+       0.216178 ms.   0.237709%. FC
+     0.00268895 ms. 0.00295677%. Squeeze
+         90.942 ms in Total
+FLOP per operator type:
+        1.98431 GFLOP.    98.9343%. Conv
+      0.0177039 GFLOP.   0.882686%. Mul
+       0.002817 GFLOP.   0.140451%. FC
+    0.000853984 GFLOP.  0.0425782%. Add
+        2.00568 GFLOP in Total
+Feature Memory Read per operator type:
+        120.609 MB.    54.9637%. Mul
+        86.3512 MB.    39.3519%. Conv
+        6.83187 MB.    3.11341%. Add
+        5.64163 MB.      2.571%. FC
+        219.433 MB in Total
+Feature Memory Written per operator type:
+        70.8155 MB.    54.6573%. Mul
+        55.3273 MB.    42.7031%. Conv
+        3.41594 MB.    2.63651%. Add
+          0.004 MB. 0.00308731%. FC
+        129.563 MB in Total
+Parameter Memory per operator type:
+        30.4721 MB.    84.3913%. Conv
+          5.636 MB.    15.6087%. FC
+              0 MB.          0%. Add
+              0 MB.          0%. Mul
+        36.1081 MB in Total
+```
+
+## MixNet-M
+### Optimized
+```
+Main run finished. Milliseconds per iter: 63.1122. Iters per second: 15.8448
+Time per operator type:
+        48.1139 ms.    75.2052%. Conv
+         7.1341 ms.    11.1511%. Sigmoid
+        2.63706 ms.    4.12189%. SpatialBN
+        1.73186 ms.    2.70701%. Mul
+        1.38707 ms.    2.16809%. Split
+        1.29322 ms.    2.02139%. Concat
+        1.00093 ms.    1.56452%. Relu
+       0.235309 ms.   0.367803%. Add
+       0.221579 ms.   0.346343%. FC
+       0.219315 ms.   0.342803%. AveragePool
+     0.00250145 ms. 0.00390993%. Squeeze
+        63.9768 ms in Total
+FLOP per operator type:
+       0.675273 GFLOP.    95.5827%. Conv
+      0.0221072 GFLOP.    3.12921%. SpatialBN
+     0.00538445 GFLOP.   0.762152%. Mul
+       0.003073 GFLOP.   0.434973%. FC
+    0.000642488 GFLOP.  0.0909421%. Add
+              0 GFLOP.          0%. Concat
+              0 GFLOP.          0%. Relu
+        0.70648 GFLOP in Total
+Feature Memory Read per operator type:
+        46.8424 MB.     30.502%. Conv
+        36.8626 MB.    24.0036%. Mul
+        22.3152 MB.    14.5309%. SpatialBN
+        22.1074 MB.    14.3955%. Concat
+        14.1496 MB.    9.21372%. Relu
+        6.15414 MB.    4.00735%. FC
+         5.1399 MB.    3.34692%. Add
+        153.571 MB in Total
+Feature Memory Written per operator type:
+        32.7672 MB.    28.4331%. Conv
+        22.1072 MB.    19.1831%. Concat
+        22.1072 MB.    19.1831%. SpatialBN
+        21.5378 MB.     18.689%. Mul
+        14.1496 MB.    12.2781%. Relu
+        2.56995 MB.    2.23003%. Add
+          0.004 MB. 0.00347092%. FC
+        115.243 MB in Total
+Parameter Memory per operator type:
+        13.7059 MB.     68.674%. Conv
+          6.148 MB.    30.8049%. FC
+          0.104 MB.   0.521097%. SpatialBN
+              0 MB.          0%. Add
+              0 MB.          0%. Concat
+              0 MB.          0%. Mul
+              0 MB.          0%. Relu
+        19.9579 MB in Total
+```
+
+## TF MobileNet-V3 Large 1.0
+
+### Optimized
+```
+Main run finished. Milliseconds per iter: 22.0495. Iters per second: 45.3525
+Time per operator type:
+         17.437 ms.    80.0087%. Conv
+        1.27662 ms.     5.8577%. Add
+        1.12759 ms.    5.17387%. Div
+       0.701155 ms.    3.21721%. Mul
+       0.562654 ms.    2.58171%. Relu
+       0.431144 ms.    1.97828%. Clip
+       0.156902 ms.   0.719936%. FC
+      0.0996858 ms.   0.457402%. AveragePool
+     0.00112455 ms. 0.00515993%. Flatten
+        21.7939 ms in Total
+FLOP per operator type:
+        0.43062 GFLOP.    98.1484%. Conv
+       0.002561 GFLOP.   0.583713%. FC
+     0.00210867 GFLOP.   0.480616%. Mul
+     0.00193868 GFLOP.   0.441871%. Add
+     0.00151532 GFLOP.   0.345377%. Div
+              0 GFLOP.          0%. Relu
+       0.438743 GFLOP in Total
+Feature Memory Read per operator type:
+        34.7967 MB.    43.9391%. Conv
+         14.496 MB.    18.3046%. Mul
+        9.44828 MB.    11.9307%. Add
+        9.26157 MB.    11.6949%. Relu
+         6.0614 MB.    7.65395%. Div
+        5.12912 MB.    6.47673%. FC
+         79.193 MB in Total
+Feature Memory Written per operator type:
+        17.6247 MB.    35.8656%. Conv
+        9.26157 MB.     18.847%. Relu
+        8.43469 MB.    17.1643%. Mul
+        7.75472 MB.    15.7806%. Add
+        6.06128 MB.    12.3345%. Div
+          0.004 MB. 0.00813985%. FC
+        49.1409 MB in Total
+Parameter Memory per operator type:
+        16.6851 MB.    76.5052%. Conv
+          5.124 MB.    23.4948%. FC
+              0 MB.          0%. Add
+              0 MB.          0%. Div
+              0 MB.          0%. Mul
+              0 MB.          0%. Relu
+        21.8091 MB in Total
+```
+
+## MobileNet-V3 (RW)
+
+### Unoptimized
+```
+Main run finished. Milliseconds per iter: 24.8316. Iters per second: 40.2712
+Time per operator type:
+        15.9266 ms.    69.2624%. Conv
+        2.36551 ms.    10.2873%. SpatialBN
+        1.39102 ms.    6.04936%. Add
+        1.30327 ms.    5.66773%. Div
+       0.737014 ms.    3.20517%. Mul
+       0.639697 ms.    2.78195%. Relu
+       0.375681 ms.    1.63378%. Clip
+       0.153126 ms.   0.665921%. FC
+      0.0993787 ms.   0.432184%. AveragePool
+      0.0032632 ms.  0.0141912%. Squeeze
+        22.9946 ms in Total
+FLOP per operator type:
+       0.430616 GFLOP.    94.4041%. Conv
+      0.0175992 GFLOP.    3.85829%. SpatialBN
+       0.002561 GFLOP.   0.561449%. FC
+     0.00210961 GFLOP.    0.46249%. Mul
+     0.00173891 GFLOP.   0.381223%. Add
+     0.00151626 GFLOP.    0.33241%. Div
+              0 GFLOP.          0%. Relu
+       0.456141 GFLOP in Total
+Feature Memory Read per operator type:
+        34.7354 MB.    36.4363%. Conv
+        17.7944 MB.    18.6658%. SpatialBN
+        14.5035 MB.    15.2137%. Mul
+        9.25778 MB.    9.71113%. Relu
+        7.84641 MB.    8.23064%. Add
+        6.06516 MB.    6.36216%. Div
+        5.12912 MB.    5.38029%. FC
+        95.3317 MB in Total
+Feature Memory Written per operator type:
+        17.6246 MB.    26.7264%. Conv
+        17.5992 MB.    26.6878%. SpatialBN
+        9.25778 MB.    14.0387%. Relu
+        8.43843 MB.    12.7962%. Mul
+        6.95565 MB.    10.5477%. Add
+        6.06502 MB.    9.19713%. Div
+          0.004 MB. 0.00606568%. FC
+        65.9447 MB in Total
+Parameter Memory per operator type:
+        16.6778 MB.    76.1564%. Conv
+          5.124 MB.    23.3979%. FC
+         0.0976 MB.   0.445674%. SpatialBN
+              0 MB.          0%. Add
+              0 MB.          0%. Div
+              0 MB.          0%. Mul
+              0 MB.          0%. Relu
+        21.8994 MB in Total
+
+```
+### Optimized
+
+```
+Main run finished. Milliseconds per iter: 22.0981. Iters per second: 45.2527
+Time per operator type:
+         17.146 ms.    78.8965%. Conv
+        1.38453 ms.    6.37084%. Add
+        1.30991 ms.    6.02749%. Div
+       0.685417 ms.    3.15391%. Mul
+       0.532589 ms.    2.45068%. Relu
+       0.418263 ms.    1.92461%. Clip
+        0.15128 ms.   0.696106%. FC
+       0.102065 ms.   0.469648%. AveragePool
+      0.0022143 ms.   0.010189%. Squeeze
+        21.7323 ms in Total
+FLOP per operator type:
+       0.430616 GFLOP.    98.1927%. Conv
+       0.002561 GFLOP.   0.583981%. FC
+     0.00210961 GFLOP.   0.481051%. Mul
+     0.00173891 GFLOP.   0.396522%. Add
+     0.00151626 GFLOP.    0.34575%. Div
+              0 GFLOP.          0%. Relu
+       0.438542 GFLOP in Total
+Feature Memory Read per operator type:
+        34.7842 MB.     44.833%. Conv
+        14.5035 MB.    18.6934%. Mul
+        9.25778 MB.    11.9323%. Relu
+        7.84641 MB.    10.1132%. Add
+        6.06516 MB.    7.81733%. Div
+        5.12912 MB.    6.61087%. FC
+        77.5861 MB in Total
+Feature Memory Written per operator type:
+        17.6246 MB.    36.4556%. Conv
+        9.25778 MB.    19.1492%. Relu
+        8.43843 MB.    17.4544%. Mul
+        6.95565 MB.    14.3874%. Add
+        6.06502 MB.    12.5452%. Div
+          0.004 MB. 0.00827378%. FC
+        48.3455 MB in Total
+Parameter Memory per operator type:
+        16.6778 MB.    76.4973%. Conv
+          5.124 MB.    23.5027%. FC
+              0 MB.          0%. Add
+              0 MB.          0%. Div
+              0 MB.          0%. Mul
+              0 MB.          0%. Relu
+        21.8018 MB in Total
+
+```
+
+## MnasNet-A1
+
+### Unoptimized
+```
+Main run finished. Milliseconds per iter: 30.0892. Iters per second: 33.2345
+Time per operator type:
+        24.4656 ms.    79.0905%. Conv
+        4.14958 ms.    13.4144%. SpatialBN
+        1.60598 ms.    5.19169%. Relu
+       0.295219 ms.    0.95436%. Mul
+       0.187609 ms.   0.606486%. FC
+       0.120556 ms.   0.389724%. AveragePool
+        0.09036 ms.   0.292109%. Add
+       0.015727 ms.   0.050841%. Sigmoid
+     0.00306205 ms. 0.00989875%. Squeeze
+        30.9337 ms in Total
+FLOP per operator type:
+       0.620598 GFLOP.    95.6434%. Conv
+      0.0248873 GFLOP.     3.8355%. SpatialBN
+       0.002561 GFLOP.   0.394688%. FC
+    0.000597408 GFLOP.  0.0920695%. Mul
+    0.000222656 GFLOP.  0.0343146%. Add
+              0 GFLOP.          0%. Relu
+       0.648867 GFLOP in Total
+Feature Memory Read per operator type:
+        35.5457 MB.    38.4109%. Conv
+        25.1552 MB.    27.1829%. SpatialBN
+        22.5235 MB.     24.339%. Relu
+        5.12912 MB.    5.54256%. FC
+        2.40586 MB.    2.59978%. Mul
+        1.78125 MB.    1.92483%. Add
+        92.5406 MB in Total
+Feature Memory Written per operator type:
+        24.9042 MB.    32.9424%. Conv
+        24.8873 MB.      32.92%. SpatialBN
+        22.5235 MB.    29.7932%. Relu
+        2.38963 MB.    3.16092%. Mul
+       0.890624 MB.    1.17809%. Add
+          0.004 MB. 0.00529106%. FC
+        75.5993 MB in Total
+Parameter Memory per operator type:
+        10.2732 MB.    66.1459%. Conv
+          5.124 MB.    32.9917%. FC
+       0.133952 MB.    0.86247%. SpatialBN
+              0 MB.          0%. Add
+              0 MB.          0%. Mul
+              0 MB.          0%. Relu
+        15.5312 MB in Total
+```
+
+### Optimized
+```
+Main run finished. Milliseconds per iter: 24.2367. Iters per second: 41.2597
+Time per operator type:
+        22.0547 ms.    91.1375%. Conv
+        1.49096 ms.    6.16116%. Relu
+       0.253417 ms.     1.0472%. Mul
+        0.18506 ms.    0.76473%. FC
+       0.112942 ms.   0.466717%. AveragePool
+       0.086769 ms.   0.358559%. Add
+      0.0127889 ms.  0.0528479%. Sigmoid
+      0.0027346 ms.  0.0113003%. Squeeze
+        24.1994 ms in Total
+FLOP per operator type:
+       0.620598 GFLOP.    99.4581%. Conv
+       0.002561 GFLOP.    0.41043%. FC
+    0.000597408 GFLOP.  0.0957417%. Mul
+    0.000222656 GFLOP.  0.0356832%. Add
+              0 GFLOP.          0%. Relu
+       0.623979 GFLOP in Total
+Feature Memory Read per operator type:
+        35.6127 MB.    52.7968%. Conv
+        22.5235 MB.    33.3917%. Relu
+        5.12912 MB.    7.60406%. FC
+        2.40586 MB.    3.56675%. Mul
+        1.78125 MB.    2.64075%. Add
+        67.4524 MB in Total
+Feature Memory Written per operator type:
+        24.9042 MB.    49.1092%. Conv
+        22.5235 MB.    44.4145%. Relu
+        2.38963 MB.    4.71216%. Mul
+       0.890624 MB.    1.75624%. Add
+          0.004 MB. 0.00788768%. FC
+         50.712 MB in Total
+Parameter Memory per operator type:
+        10.2732 MB.    66.7213%. Conv
+          5.124 MB.    33.2787%. FC
+              0 MB.          0%. Add
+              0 MB.          0%. Mul
+              0 MB.          0%. Relu
+        15.3972 MB in Total
+```
+## MnasNet-B1
+
+### Unoptimized
+```
+Main run finished. Milliseconds per iter: 28.3109. Iters per second: 35.322
+Time per operator type:
+        29.1121 ms.    83.3081%. Conv
+        4.14959 ms.    11.8746%. SpatialBN
+        1.35823 ms.    3.88675%. Relu
+       0.186188 ms.   0.532802%. FC
+       0.116244 ms.   0.332647%. Add
+       0.018641 ms.  0.0533437%. AveragePool
+      0.0040904 ms.  0.0117052%. Squeeze
+        34.9451 ms in Total
+FLOP per operator type:
+       0.626272 GFLOP.    96.2088%. Conv
+      0.0218266 GFLOP.    3.35303%. SpatialBN
+       0.002561 GFLOP.   0.393424%. FC
+    0.000291648 GFLOP.  0.0448034%. Add
+              0 GFLOP.          0%. Relu
+       0.650951 GFLOP in Total
+Feature Memory Read per operator type:
+        34.4354 MB.    41.3788%. Conv
+        22.1299 MB.    26.5921%. SpatialBN
+        19.1923 MB.    23.0622%. Relu
+        5.12912 MB.    6.16333%. FC
+        2.33318 MB.    2.80364%. Add
+        83.2199 MB in Total
+Feature Memory Written per operator type:
+        21.8266 MB.    34.0955%. Conv
+        21.8266 MB.    34.0955%. SpatialBN
+        19.1923 MB.    29.9805%. Relu
+        1.16659 MB.    1.82234%. Add
+          0.004 MB. 0.00624844%. FC
+         64.016 MB in Total
+Parameter Memory per operator type:
+        12.2576 MB.    69.9104%. Conv
+          5.124 MB.    29.2245%. FC
+        0.15168 MB.   0.865099%. SpatialBN
+              0 MB.          0%. Add
+              0 MB.          0%. Relu
+        17.5332 MB in Total
+```
+
+### Optimized
+```
+Main run finished. Milliseconds per iter: 26.6364. Iters per second: 37.5426
+Time per operator type:
+        24.9888 ms.    94.0962%. Conv
+        1.26147 ms.    4.75011%. Relu
+       0.176234 ms.   0.663619%. FC
+       0.113309 ms.   0.426672%. Add
+      0.0138708 ms.  0.0522311%. AveragePool
+     0.00295685 ms.  0.0111341%. Squeeze
+        26.5566 ms in Total
+FLOP per operator type:
+       0.626272 GFLOP.    99.5466%. Conv
+       0.002561 GFLOP.   0.407074%. FC
+    0.000291648 GFLOP.  0.0463578%. Add
+              0 GFLOP.          0%. Relu
+       0.629124 GFLOP in Total
+Feature Memory Read per operator type:
+        34.5112 MB.    56.4224%. Conv
+        19.1923 MB.    31.3775%. Relu
+        5.12912 MB.     8.3856%. FC
+        2.33318 MB.    3.81452%. Add
+        61.1658 MB in Total
+Feature Memory Written per operator type:
+        21.8266 MB.    51.7346%. Conv
+        19.1923 MB.    45.4908%. Relu
+        1.16659 MB.    2.76513%. Add
+          0.004 MB. 0.00948104%. FC
+        42.1895 MB in Total
+Parameter Memory per operator type:
+        12.2576 MB.    70.5205%. Conv
+          5.124 MB.    29.4795%. FC
+              0 MB.          0%. Add
+              0 MB.          0%. Relu
+        17.3816 MB in Total
+```

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "{}"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2020 Ross Wightman
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/README.md

@@ -0,0 +1,323 @@
+# (Generic) EfficientNets for PyTorch
+
+A 'generic' implementation of EfficientNet, MixNet, MobileNetV3, etc. that covers most of the compute/parameter efficient architectures derived from the MobileNet V1/V2 block sequence, including those found via automated neural architecture search. 
+
+All models are implemented by GenEfficientNet or MobileNetV3 classes, with string based architecture definitions to configure the block layouts (idea from [here](https://github.com/tensorflow/tpu/blob/master/models/official/mnasnet/mnasnet_models.py))
+
+## What's New
+
+### Aug 19, 2020
+* Add updated PyTorch trained EfficientNet-B3 weights trained by myself with `timm` (82.1 top-1)
+* Add PyTorch trained EfficientNet-Lite0 contributed by [@hal-314](https://github.com/hal-314) (75.5 top-1)
+* Update ONNX and Caffe2 export / utility scripts to work with latest PyTorch / ONNX
+* ONNX runtime based validation script added
+* activations (mostly) brought in sync with `timm` equivalents
+
+
+### April 5, 2020
+* Add some newly trained MobileNet-V2 models trained with latest h-params, rand augment. They compare quite favourably to EfficientNet-Lite
+  * 3.5M param MobileNet-V2 100 @ 73%
+  * 4.5M param MobileNet-V2 110d @ 75%
+  * 6.1M param MobileNet-V2 140 @ 76.5%
+  * 5.8M param MobileNet-V2 120d @ 77.3%
+  
+### March 23, 2020
+ * Add EfficientNet-Lite models w/ weights ported from [Tensorflow TPU](https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet/lite)
+ * Add PyTorch trained MobileNet-V3 Large weights with 75.77% top-1
+ * IMPORTANT CHANGE (if training from scratch) - weight init changed to better match Tensorflow impl, set `fix_group_fanout=False` in `initialize_weight_goog` for old behavior
+
+### Feb 12, 2020
+ * Add EfficientNet-L2 and B0-B7 NoisyStudent weights ported from [Tensorflow TPU](https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet)
+ * Port new EfficientNet-B8 (RandAugment) weights from TF TPU, these are different than the B8 AdvProp, different input normalization.
+ * Add RandAugment PyTorch trained EfficientNet-ES (EdgeTPU-Small) weights with 78.1 top-1. Trained by [Andrew Lavin](https://github.com/andravin)
+
+### Jan 22, 2020
+ * Update weights for EfficientNet B0, B2, B3 and MixNet-XL with latest RandAugment trained weights. Trained with (https://github.com/rwightman/pytorch-image-models)
+ * Fix torchscript compatibility for PyTorch 1.4, add torchscript support for MixedConv2d using ModuleDict
+ * Test models, torchscript, onnx export with PyTorch 1.4 -- no issues
+
+### Nov 22, 2019
+ * New top-1 high! Ported official TF EfficientNet AdvProp (https://arxiv.org/abs/1911.09665) weights and B8 model spec. Created a new set of `ap` models since they use a different
+ preprocessing (Inception mean/std) from the original EfficientNet base/AA/RA weights.
+ 
+### Nov 15, 2019
+ * Ported official TF MobileNet-V3 float32 large/small/minimalistic weights
+ * Modifications to MobileNet-V3 model and components to support some additional config needed for differences between TF MobileNet-V3 and mine
+
+### Oct 30, 2019
+ * Many of the models will now work with torch.jit.script, MixNet being the biggest exception
+ * Improved interface for enabling torchscript or ONNX export compatible modes (via config)
+ * Add JIT optimized mem-efficient Swish/Mish autograd.fn in addition to memory-efficient autgrad.fn
+ * Activation factory to select best version of activation by name or override one globally
+ * Add pretrained checkpoint load helper that handles input conv and classifier changes
+ 
+### Oct 27, 2019
+ * Add CondConv EfficientNet variants ported from https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet/condconv
+ * Add RandAug weights for TF EfficientNet B5 and B7 from https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet
+ * Bring over MixNet-XL model and depth scaling algo from my pytorch-image-models code base
+ * Switch activations and global pooling to modules
+ * Add memory-efficient Swish/Mish impl
+ * Add as_sequential() method to all models and allow as an argument in entrypoint fns
+ * Move MobileNetV3 into own file since it has a different head
+ * Remove ChamNet, MobileNet V2/V1 since they will likely never be used here
+
+## Models
+
+Implemented models include:
+  * EfficientNet NoisyStudent (B0-B7, L2) (https://arxiv.org/abs/1911.04252)
+  * EfficientNet AdvProp (B0-B8) (https://arxiv.org/abs/1911.09665)
+  * EfficientNet (B0-B8) (https://arxiv.org/abs/1905.11946)
+  * EfficientNet-EdgeTPU (S, M, L) (https://ai.googleblog.com/2019/08/efficientnet-edgetpu-creating.html)
+  * EfficientNet-CondConv (https://arxiv.org/abs/1904.04971)
+  * EfficientNet-Lite (https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet/lite)
+  * MixNet (https://arxiv.org/abs/1907.09595)
+  * MNASNet B1, A1 (Squeeze-Excite), and Small (https://arxiv.org/abs/1807.11626)
+  * MobileNet-V3 (https://arxiv.org/abs/1905.02244)
+  * FBNet-C (https://arxiv.org/abs/1812.03443)
+  * Single-Path NAS (https://arxiv.org/abs/1904.02877)
+    
+I originally implemented and trained some these models with code [here](https://github.com/rwightman/pytorch-image-models), this repository contains just the GenEfficientNet models, validation, and associated ONNX/Caffe2 export code. 
+
+## Pretrained
+
+I've managed to train several of the models to accuracies close to or above the originating papers and official impl. My training code is here: https://github.com/rwightman/pytorch-image-models
+
+
+|Model | Prec@1 (Err) | Prec@5 (Err) | Param#(M) | MAdds(M) | Image Scaling | Resolution | Crop |
+|---|---|---|---|---|---|---|---|
+| efficientnet_b3 | 82.240 (17.760) | 96.116 (3.884) | 12.23 | TBD | bicubic | 320 | 1.0 |
+| efficientnet_b3 | 82.076 (17.924) | 96.020 (3.980) | 12.23 | TBD | bicubic | 300 | 0.904 |
+| mixnet_xl | 81.074 (18.926) | 95.282 (4.718) | 11.90 | TBD | bicubic | 256 | 1.0 |
+| efficientnet_b2 | 80.612 (19.388) | 95.318 (4.682) | 9.1 | TBD | bicubic | 288 | 1.0 |
+| mixnet_xl | 80.476 (19.524) | 94.936 (5.064) | 11.90 | TBD | bicubic | 224 | 0.875 |
+| efficientnet_b2 | 80.288 (19.712) | 95.166 (4.834) | 9.1 | 1003 | bicubic | 260 | 0.890 |
+| mixnet_l | 78.976 (21.024 | 94.184 (5.816) | 7.33 | TBD | bicubic | 224 | 0.875 |
+| efficientnet_b1 | 78.692 (21.308) | 94.086 (5.914) | 7.8 | 694 | bicubic | 240 | 0.882 |
+| efficientnet_es | 78.066 (21.934) | 93.926 (6.074) | 5.44 | TBD | bicubic | 224 | 0.875 |
+| efficientnet_b0 | 77.698 (22.302) | 93.532 (6.468) | 5.3 | 390 | bicubic | 224 | 0.875 |
+| mobilenetv2_120d | 77.294 (22.706 | 93.502 (6.498) | 5.8 | TBD | bicubic | 224 | 0.875 |
+| mixnet_m | 77.256 (22.744) | 93.418 (6.582) | 5.01 | 353 | bicubic | 224 | 0.875 |
+| mobilenetv2_140 | 76.524 (23.476) | 92.990 (7.010) | 6.1 | TBD | bicubic | 224 | 0.875 |
+| mixnet_s | 75.988 (24.012) | 92.794 (7.206) | 4.13 | TBD | bicubic | 224 | 0.875 |
+| mobilenetv3_large_100 | 75.766 (24.234) | 92.542 (7.458) | 5.5 | TBD | bicubic | 224 | 0.875 |
+| mobilenetv3_rw | 75.634 (24.366) | 92.708 (7.292) | 5.5 | 219 | bicubic | 224 | 0.875 |
+| efficientnet_lite0 | 75.472 (24.528) | 92.520 (7.480) | 4.65 | TBD | bicubic | 224 | 0.875 |
+| mnasnet_a1 | 75.448 (24.552) | 92.604 (7.396) | 3.9 | 312 | bicubic | 224 | 0.875 |
+| fbnetc_100 | 75.124 (24.876) | 92.386 (7.614) | 5.6 | 385 | bilinear | 224 | 0.875 |
+| mobilenetv2_110d | 75.052 (24.948) | 92.180 (7.820) | 4.5 | TBD | bicubic | 224 | 0.875 |
+| mnasnet_b1 | 74.658 (25.342) | 92.114 (7.886) | 4.4 | 315 | bicubic | 224 | 0.875 |
+| spnasnet_100 | 74.084 (25.916)  | 91.818 (8.182) | 4.4 | TBD | bilinear | 224 | 0.875 |
+| mobilenetv2_100 | 72.978 (27.022) | 91.016 (8.984) | 3.5 | TBD | bicubic | 224 | 0.875 |
+
+
+More pretrained models to come...
+
+
+## Ported Weights
+
+The weights ported from Tensorflow checkpoints for the EfficientNet models do pretty much match accuracy in Tensorflow once a SAME convolution padding equivalent is added, and the same crop factors, image scaling, etc (see table) are used via cmd line args.
+
+**IMPORTANT:** 
+* Tensorflow ported weights for EfficientNet AdvProp (AP), EfficientNet EdgeTPU, EfficientNet-CondConv, EfficientNet-Lite, and MobileNet-V3 models use Inception style (0.5, 0.5, 0.5) for mean and std.
+* Enabling the Tensorflow preprocessing pipeline with `--tf-preprocessing` at validation time will improve scores by 0.1-0.5%, very close to original TF impl.
+
+To run validation for tf_efficientnet_b5:
+`python validate.py /path/to/imagenet/validation/ --model tf_efficientnet_b5 -b 64 --img-size 456 --crop-pct 0.934 --interpolation bicubic`
+
+To run validation w/ TF preprocessing for tf_efficientnet_b5:
+`python validate.py /path/to/imagenet/validation/ --model tf_efficientnet_b5 -b 64 --img-size 456 --tf-preprocessing`
+
+To run validation for a model with Inception preprocessing, ie EfficientNet-B8 AdvProp:
+`python validate.py /path/to/imagenet/validation/ --model tf_efficientnet_b8_ap -b 48 --num-gpu 2 --img-size 672 --crop-pct 0.954 --mean 0.5 --std 0.5`
+
+|Model | Prec@1 (Err) | Prec@5 (Err) | Param # | Image Scaling  | Image Size | Crop | 
+|---|---|---|---|---|---|---|
+| tf_efficientnet_l2_ns *tfp | 88.352 (11.648) | 98.652 (1.348) | 480 | bicubic | 800 | N/A |
+| tf_efficientnet_l2_ns      | TBD | TBD | 480 | bicubic | 800 | 0.961 |
+| tf_efficientnet_l2_ns_475      | 88.234 (11.766) | 98.546 (1.454) | 480 | bicubic | 475 | 0.936 |
+| tf_efficientnet_l2_ns_475 *tfp | 88.172 (11.828) | 98.566 (1.434) | 480 | bicubic | 475 | N/A |
+| tf_efficientnet_b7_ns *tfp | 86.844 (13.156) | 98.084 (1.916) | 66.35 | bicubic | 600 | N/A |
+| tf_efficientnet_b7_ns      | 86.840 (13.160) | 98.094 (1.906) | 66.35 | bicubic | 600 | N/A |
+| tf_efficientnet_b6_ns      | 86.452 (13.548) | 97.882 (2.118) | 43.04 | bicubic | 528 | N/A |
+| tf_efficientnet_b6_ns *tfp | 86.444 (13.556) | 97.880 (2.120) | 43.04 | bicubic | 528 | N/A |
+| tf_efficientnet_b5_ns *tfp | 86.064 (13.936) | 97.746 (2.254) | 30.39 | bicubic | 456 | N/A |
+| tf_efficientnet_b5_ns      | 86.088 (13.912) | 97.752 (2.248) | 30.39 | bicubic | 456 | N/A |
+| tf_efficientnet_b8_ap *tfp | 85.436 (14.564) | 97.272 (2.728) | 87.4 | bicubic | 672 | N/A |
+| tf_efficientnet_b8 *tfp    | 85.384 (14.616) | 97.394 (2.606) | 87.4 | bicubic | 672 | N/A |
+| tf_efficientnet_b8         | 85.370 (14.630) | 97.390 (2.610) | 87.4 | bicubic | 672 | 0.954 |
+| tf_efficientnet_b8_ap      | 85.368 (14.632) | 97.294 (2.706) | 87.4 | bicubic | 672 | 0.954 |
+| tf_efficientnet_b4_ns *tfp | 85.298 (14.702) | 97.504 (2.496) | 19.34 | bicubic | 380 | N/A |
+| tf_efficientnet_b4_ns      | 85.162 (14.838) | 97.470 (2.530) | 19.34 | bicubic | 380 | 0.922 |
+| tf_efficientnet_b7_ap *tfp | 85.154 (14.846) | 97.244 (2.756) | 66.35 | bicubic | 600 | N/A |
+| tf_efficientnet_b7_ap      | 85.118 (14.882) | 97.252 (2.748) | 66.35 | bicubic | 600 | 0.949 |
+| tf_efficientnet_b7 *tfp | 84.940 (15.060) | 97.214 (2.786) | 66.35 | bicubic | 600 | N/A |
+| tf_efficientnet_b7      | 84.932 (15.068) | 97.208 (2.792) | 66.35 | bicubic | 600 | 0.949 |
+| tf_efficientnet_b6_ap      | 84.786 (15.214) | 97.138 (2.862) | 43.04 | bicubic | 528 | 0.942 |
+| tf_efficientnet_b6_ap *tfp | 84.760 (15.240) | 97.124 (2.876) | 43.04 | bicubic | 528 | N/A |
+| tf_efficientnet_b5_ap *tfp | 84.276 (15.724) | 96.932 (3.068) | 30.39 | bicubic | 456 | N/A |
+| tf_efficientnet_b5_ap      | 84.254 (15.746) | 96.976 (3.024) | 30.39 | bicubic | 456 | 0.934 |
+| tf_efficientnet_b6 *tfp  | 84.140 (15.860) | 96.852 (3.148) | 43.04 | bicubic | 528 | N/A |
+| tf_efficientnet_b6       | 84.110 (15.890) | 96.886 (3.114) | 43.04 | bicubic | 528 | 0.942 |
+| tf_efficientnet_b3_ns *tfp | 84.054 (15.946) | 96.918 (3.082) | 12.23 | bicubic | 300 | N/A |
+| tf_efficientnet_b3_ns      | 84.048 (15.952) | 96.910 (3.090) | 12.23 | bicubic | 300 | .904 |
+| tf_efficientnet_b5 *tfp  | 83.822 (16.178) | 96.756 (3.244) | 30.39 | bicubic | 456 | N/A |
+| tf_efficientnet_b5       | 83.812 (16.188) | 96.748 (3.252) | 30.39 | bicubic | 456 | 0.934 |
+| tf_efficientnet_b4_ap *tfp | 83.278 (16.722) | 96.376 (3.624) | 19.34 | bicubic | 380 | N/A |
+| tf_efficientnet_b4_ap      | 83.248 (16.752) | 96.388 (3.612) | 19.34 | bicubic | 380 | 0.922 |
+| tf_efficientnet_b4       | 83.022 (16.978) | 96.300 (3.700) | 19.34 | bicubic | 380 | 0.922 |
+| tf_efficientnet_b4 *tfp  | 82.948 (17.052) | 96.308 (3.692) | 19.34 | bicubic | 380 | N/A |
+| tf_efficientnet_b2_ns *tfp | 82.436 (17.564) | 96.268 (3.732) | 9.11 | bicubic | 260 | N/A |
+| tf_efficientnet_b2_ns      | 82.380 (17.620) | 96.248 (3.752) | 9.11 | bicubic | 260 | 0.89 |
+| tf_efficientnet_b3_ap *tfp | 81.882 (18.118) | 95.662 (4.338) | 12.23 | bicubic | 300 | N/A |
+| tf_efficientnet_b3_ap      | 81.828 (18.172) | 95.624 (4.376) | 12.23 | bicubic | 300 | 0.904 |
+| tf_efficientnet_b3       | 81.636 (18.364) | 95.718 (4.282) | 12.23 | bicubic | 300 | 0.904 |
+| tf_efficientnet_b3 *tfp  | 81.576 (18.424) | 95.662 (4.338) | 12.23 | bicubic | 300 | N/A |
+| tf_efficientnet_lite4      | 81.528 (18.472) | 95.668 (4.332) | 13.00  | bilinear | 380 | 0.92 |
+| tf_efficientnet_b1_ns *tfp | 81.514 (18.486) | 95.776 (4.224) | 7.79 | bicubic | 240 | N/A |
+| tf_efficientnet_lite4 *tfp | 81.502 (18.498) | 95.676 (4.324) | 13.00  | bilinear | 380 | N/A |
+| tf_efficientnet_b1_ns      | 81.388 (18.612) | 95.738 (4.262) | 7.79 | bicubic | 240 | 0.88 |
+| tf_efficientnet_el       | 80.534 (19.466) | 95.190 (4.810) | 10.59 | bicubic | 300 | 0.904 |
+| tf_efficientnet_el *tfp  | 80.476 (19.524) | 95.200 (4.800) | 10.59 | bicubic | 300 | N/A |
+| tf_efficientnet_b2_ap *tfp | 80.420 (19.580) | 95.040 (4.960) | 9.11 | bicubic | 260 | N/A |
+| tf_efficientnet_b2_ap    | 80.306 (19.694) | 95.028 (4.972) | 9.11 | bicubic | 260 | 0.890 |
+| tf_efficientnet_b2 *tfp  | 80.188 (19.812) | 94.974 (5.026) | 9.11 | bicubic | 260 | N/A |
+| tf_efficientnet_b2       | 80.086 (19.914) | 94.908 (5.092) | 9.11 | bicubic | 260 | 0.890 |
+| tf_efficientnet_lite3       | 79.812 (20.188) | 94.914 (5.086) | 8.20  | bilinear | 300 | 0.904 |
+| tf_efficientnet_lite3 *tfp  | 79.734 (20.266) | 94.838 (5.162) | 8.20  | bilinear | 300 | N/A |
+| tf_efficientnet_b1_ap *tfp | 79.532 (20.468) | 94.378 (5.622) | 7.79 | bicubic | 240 | N/A |
+| tf_efficientnet_cc_b1_8e *tfp | 79.464 (20.536)| 94.492 (5.508) | 39.7 | bicubic | 240 | 0.88 |
+| tf_efficientnet_cc_b1_8e | 79.298 (20.702) | 94.364 (5.636) | 39.7 | bicubic | 240 | 0.88 |
+| tf_efficientnet_b1_ap    | 79.278 (20.722) | 94.308 (5.692) | 7.79 | bicubic | 240 | 0.88 |
+| tf_efficientnet_b1 *tfp  | 79.172 (20.828) | 94.450 (5.550) | 7.79 | bicubic | 240 | N/A |
+| tf_efficientnet_em *tfp  | 78.958 (21.042) | 94.458 (5.542) | 6.90 | bicubic | 240 | N/A |
+| tf_efficientnet_b0_ns *tfp | 78.806 (21.194) | 94.496 (5.504) | 5.29 | bicubic | 224 | N/A |
+| tf_mixnet_l *tfp         | 78.846 (21.154) | 94.212 (5.788) | 7.33 | bilinear | 224 | N/A |
+| tf_efficientnet_b1       | 78.826 (21.174) | 94.198 (5.802) | 7.79 | bicubic | 240 | 0.88 |
+| tf_mixnet_l              | 78.770 (21.230) | 94.004 (5.996) | 7.33 | bicubic | 224 | 0.875 |
+| tf_efficientnet_em       | 78.742 (21.258) | 94.332 (5.668) | 6.90 | bicubic | 240 | 0.875 |
+| tf_efficientnet_b0_ns    | 78.658 (21.342) | 94.376 (5.624) | 5.29 | bicubic | 224 | 0.875 |
+| tf_efficientnet_cc_b0_8e *tfp | 78.314 (21.686) | 93.790 (6.210) | 24.0 | bicubic | 224 | 0.875 |
+| tf_efficientnet_cc_b0_8e | 77.908 (22.092) | 93.656 (6.344) | 24.0 | bicubic | 224 | 0.875 |
+| tf_efficientnet_cc_b0_4e *tfp | 77.746 (22.254) | 93.552 (6.448) | 13.3 | bicubic | 224 | 0.875 |
+| tf_efficientnet_cc_b0_4e | 77.304 (22.696) | 93.332 (6.668) | 13.3 | bicubic | 224 | 0.875 |
+| tf_efficientnet_es *tfp  | 77.616 (22.384) | 93.750 (6.250) | 5.44 | bicubic | 224 | N/A |
+| tf_efficientnet_lite2 *tfp  | 77.544 (22.456) | 93.800 (6.200) | 6.09  | bilinear | 260 | N/A |
+| tf_efficientnet_lite2       | 77.460 (22.540) | 93.746 (6.254) | 6.09  | bicubic | 260 | 0.89 |
+| tf_efficientnet_b0_ap *tfp | 77.514 (22.486) | 93.576 (6.424) | 5.29  | bicubic | 224 | N/A |
+| tf_efficientnet_es       | 77.264 (22.736) | 93.600 (6.400) | 5.44 | bicubic | 224 | N/A |
+| tf_efficientnet_b0 *tfp  | 77.258 (22.742) | 93.478 (6.522) | 5.29  | bicubic | 224 | N/A |
+| tf_efficientnet_b0_ap    | 77.084 (22.916) | 93.254 (6.746) | 5.29  | bicubic | 224 | 0.875 |
+| tf_mixnet_m *tfp         | 77.072 (22.928) | 93.368 (6.632) | 5.01 | bilinear | 224 | N/A |
+| tf_mixnet_m              | 76.950 (23.050) | 93.156 (6.844) | 5.01 | bicubic | 224 | 0.875 |
+| tf_efficientnet_b0       | 76.848 (23.152) | 93.228 (6.772) | 5.29  | bicubic | 224 | 0.875 |
+| tf_efficientnet_lite1 *tfp  | 76.764 (23.236) | 93.326 (6.674) | 5.42  | bilinear | 240 | N/A |
+| tf_efficientnet_lite1       | 76.638 (23.362) | 93.232 (6.768) | 5.42  | bicubic | 240 | 0.882 |
+| tf_mixnet_s *tfp         | 75.800 (24.200) | 92.788 (7.212) | 4.13 | bilinear | 224 | N/A |
+| tf_mobilenetv3_large_100 *tfp | 75.768 (24.232) | 92.710 (7.290) | 5.48 | bilinear | 224 | N/A |
+| tf_mixnet_s              | 75.648 (24.352) | 92.636 (7.364) | 4.13 | bicubic | 224 | 0.875 |
+| tf_mobilenetv3_large_100 | 75.516 (24.484) | 92.600 (7.400) | 5.48 | bilinear | 224 | 0.875 |
+| tf_efficientnet_lite0 *tfp  | 75.074 (24.926) | 92.314 (7.686) | 4.65  | bilinear | 224 | N/A |
+| tf_efficientnet_lite0       | 74.842 (25.158) | 92.170 (7.830) | 4.65  | bicubic | 224 | 0.875 |
+| tf_mobilenetv3_large_075 *tfp | 73.730 (26.270) | 91.616 (8.384) | 3.99 | bilinear | 224 |N/A |
+| tf_mobilenetv3_large_075 | 73.442 (26.558) | 91.352 (8.648) | 3.99 | bilinear | 224 | 0.875 |
+| tf_mobilenetv3_large_minimal_100 *tfp | 72.678 (27.322) | 90.860 (9.140) | 3.92 | bilinear | 224 | N/A |
+| tf_mobilenetv3_large_minimal_100 | 72.244 (27.756) | 90.636 (9.364) | 3.92 | bilinear | 224 | 0.875 |
+| tf_mobilenetv3_small_100 *tfp | 67.918 (32.082) | 87.958 (12.042 | 2.54 | bilinear | 224 | N/A |
+| tf_mobilenetv3_small_100 | 67.918 (32.082) | 87.662 (12.338) | 2.54 | bilinear | 224 | 0.875 |
+| tf_mobilenetv3_small_075 *tfp | 66.142 (33.858) | 86.498 (13.502) | 2.04 | bilinear | 224 | N/A |
+| tf_mobilenetv3_small_075 | 65.718 (34.282) | 86.136 (13.864) | 2.04 | bilinear | 224 | 0.875 |
+| tf_mobilenetv3_small_minimal_100 *tfp | 63.378 (36.622) | 84.802 (15.198) | 2.04 | bilinear | 224 | N/A |
+| tf_mobilenetv3_small_minimal_100 | 62.898 (37.102) | 84.230 (15.770) | 2.04 | bilinear | 224 | 0.875 |
+
+
+*tfp models validated with `tf-preprocessing` pipeline
+
+Google tf and tflite weights ported from official Tensorflow repositories
+* https://github.com/tensorflow/tpu/tree/master/models/official/mnasnet
+* https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet
+* https://github.com/tensorflow/models/tree/master/research/slim/nets/mobilenet
+
+## Usage
+
+### Environment
+
+All development and testing has been done in Conda Python 3 environments on Linux x86-64 systems, specifically Python 3.6.x, 3.7.x, 3.8.x.
+
+Users have reported that a Python 3 Anaconda install in Windows works. I have not verified this myself.
+
+PyTorch versions 1.4, 1.5, 1.6 have been tested with this code.
+
+I've tried to keep the dependencies minimal, the setup is as per the PyTorch default install instructions for Conda:
+```
+conda create -n torch-env
+conda activate torch-env
+conda install -c pytorch pytorch torchvision cudatoolkit=10.2
+```
+
+### PyTorch Hub
+
+Models can be accessed via the PyTorch Hub API
+
+```
+>>> torch.hub.list('rwightman/gen-efficientnet-pytorch')
+['efficientnet_b0', ...]
+>>> model = torch.hub.load('rwightman/gen-efficientnet-pytorch', 'efficientnet_b0', pretrained=True)
+>>> model.eval()
+>>> output = model(torch.randn(1,3,224,224))
+```
+
+### Pip
+This package can be installed via pip.
+
+Install (after conda env/install):
+```
+pip install geffnet
+```
+
+Eval use:
+```
+>>> import geffnet
+>>> m = geffnet.create_model('mobilenetv3_large_100', pretrained=True)
+>>> m.eval()
+```
+
+Train use:
+```
+>>> import geffnet
+>>> # models can also be created by using the entrypoint directly
+>>> m = geffnet.efficientnet_b2(pretrained=True, drop_rate=0.25, drop_connect_rate=0.2)
+>>> m.train()
+```
+
+Create in a nn.Sequential container, for fast.ai, etc:
+```
+>>> import geffnet
+>>> m = geffnet.mixnet_l(pretrained=True, drop_rate=0.25, drop_connect_rate=0.2, as_sequential=True)
+```
+
+### Exporting
+
+Scripts are included to
+* export models to ONNX (`onnx_export.py`)
+* optimized ONNX graph (`onnx_optimize.py` or `onnx_validate.py` w/ `--onnx-output-opt` arg)
+* validate with ONNX runtime  (`onnx_validate.py`)
+* convert ONNX model to Caffe2 (`onnx_to_caffe.py`)
+* validate in Caffe2 (`caffe2_validate.py`)
+* benchmark in Caffe2 w/ FLOPs, parameters output (`caffe2_benchmark.py`)
+
+As an example, to export the MobileNet-V3 pretrained model and then run an Imagenet validation:
+```
+python onnx_export.py --model mobilenetv3_large_100 ./mobilenetv3_100.onnx
+python onnx_validate.py /imagenet/validation/ --onnx-input ./mobilenetv3_100.onnx 
+```
+
+These scripts were tested to be working as of PyTorch 1.6 and ONNX 1.7 w/ ONNX runtime 1.4. Caffe2 compatible
+export now requires additional args mentioned in the export script (not needed in earlier versions).
+
+#### Export Notes
+1. The TF ported weights with the 'SAME' conv padding activated cannot be exported to ONNX unless `_EXPORTABLE` flag in `config.py` is set to True. Use `config.set_exportable(True)` as in the `onnx_export.py` script.
+2. TF ported models with 'SAME' padding will have the padding fixed at export time to the resolution used for export. Even though dynamic padding is supported in opset >= 11, I can't get it working.
+3. ONNX optimize facility doesn't work reliably in PyTorch 1.6 / ONNX 1.7. Fortunately, the onnxruntime based inference is working very well now and includes on the fly optimization.
+3. ONNX / Caffe2 export/import frequently breaks with different PyTorch and ONNX version releases. Please check their respective issue trackers before filing issues here.
+
+

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/caffe2_benchmark.py

@@ -0,0 +1,65 @@
+""" Caffe2 validation script
+
+This script runs Caffe2 benchmark on exported ONNX model.
+It is a useful tool for reporting model FLOPS.
+
+Copyright 2020 Ross Wightman
+"""
+import argparse
+from caffe2.python import core, workspace, model_helper
+from caffe2.proto import caffe2_pb2
+
+
+parser = argparse.ArgumentParser(description='Caffe2 Model Benchmark')
+parser.add_argument('--c2-prefix', default='', type=str, metavar='NAME',
+                    help='caffe2 model pb name prefix')
+parser.add_argument('--c2-init', default='', type=str, metavar='PATH',
+                    help='caffe2 model init .pb')
+parser.add_argument('--c2-predict', default='', type=str, metavar='PATH',
+                    help='caffe2 model predict .pb')
+parser.add_argument('-b', '--batch-size', default=1, type=int,
+                    metavar='N', help='mini-batch size (default: 1)')
+parser.add_argument('--img-size', default=224, type=int,
+                    metavar='N', help='Input image dimension, uses model default if empty')
+
+
+def main():
+    args = parser.parse_args()
+    args.gpu_id = 0
+    if args.c2_prefix:
+        args.c2_init = args.c2_prefix + '.init.pb'
+        args.c2_predict = args.c2_prefix + '.predict.pb'
+
+    model = model_helper.ModelHelper(name="le_net", init_params=False)
+
+    # Bring in the init net from init_net.pb
+    init_net_proto = caffe2_pb2.NetDef()
+    with open(args.c2_init, "rb") as f:
+        init_net_proto.ParseFromString(f.read())
+    model.param_init_net = core.Net(init_net_proto)
+
+    # bring in the predict net from predict_net.pb
+    predict_net_proto = caffe2_pb2.NetDef()
+    with open(args.c2_predict, "rb") as f:
+        predict_net_proto.ParseFromString(f.read())
+    model.net = core.Net(predict_net_proto)
+
+    # CUDA performance not impressive
+    #device_opts = core.DeviceOption(caffe2_pb2.PROTO_CUDA, args.gpu_id)
+    #model.net.RunAllOnGPU(gpu_id=args.gpu_id, use_cudnn=True)
+    #model.param_init_net.RunAllOnGPU(gpu_id=args.gpu_id, use_cudnn=True)
+
+    input_blob = model.net.external_inputs[0]
+    model.param_init_net.GaussianFill(
+        [],
+        input_blob.GetUnscopedName(),
+        shape=(args.batch_size, 3, args.img_size, args.img_size),
+        mean=0.0,
+        std=1.0)
+    workspace.RunNetOnce(model.param_init_net)
+    workspace.CreateNet(model.net, overwrite=True)
+    workspace.BenchmarkNet(model.net.Proto().name, 5, 20, True)
+
+
+if __name__ == '__main__':
+    main()

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/caffe2_validate.py

@@ -0,0 +1,138 @@
+""" Caffe2 validation script
+
+This script is created to verify exported ONNX models running in Caffe2
+It utilizes the same PyTorch dataloader/processing pipeline for a
+fair comparison against the originals.
+
+Copyright 2020 Ross Wightman
+"""
+import argparse
+import numpy as np
+from caffe2.python import core, workspace, model_helper
+from caffe2.proto import caffe2_pb2
+from data import create_loader, resolve_data_config, Dataset
+from utils import AverageMeter
+import time
+
+parser = argparse.ArgumentParser(description='Caffe2 ImageNet Validation')
+parser.add_argument('data', metavar='DIR',
+                    help='path to dataset')
+parser.add_argument('--c2-prefix', default='', type=str, metavar='NAME',
+                    help='caffe2 model pb name prefix')
+parser.add_argument('--c2-init', default='', type=str, metavar='PATH',
+                    help='caffe2 model init .pb')
+parser.add_argument('--c2-predict', default='', type=str, metavar='PATH',
+                    help='caffe2 model predict .pb')
+parser.add_argument('-j', '--workers', default=2, type=int, metavar='N',
+                    help='number of data loading workers (default: 2)')
+parser.add_argument('-b', '--batch-size', default=256, type=int,
+                    metavar='N', help='mini-batch size (default: 256)')
+parser.add_argument('--img-size', default=None, type=int,
+                    metavar='N', help='Input image dimension, uses model default if empty')
+parser.add_argument('--mean', type=float, nargs='+', default=None, metavar='MEAN',
+                    help='Override mean pixel value of dataset')
+parser.add_argument('--std', type=float,  nargs='+', default=None, metavar='STD',
+                    help='Override std deviation of of dataset')
+parser.add_argument('--crop-pct', type=float, default=None, metavar='PCT',
+                    help='Override default crop pct of 0.875')
+parser.add_argument('--interpolation', default='', type=str, metavar='NAME',
+                    help='Image resize interpolation type (overrides model)')
+parser.add_argument('--tf-preprocessing', dest='tf_preprocessing', action='store_true',
+                    help='use tensorflow mnasnet preporcessing')
+parser.add_argument('--print-freq', '-p', default=10, type=int,
+                    metavar='N', help='print frequency (default: 10)')
+
+
+def main():
+    args = parser.parse_args()
+    args.gpu_id = 0
+    if args.c2_prefix:
+        args.c2_init = args.c2_prefix + '.init.pb'
+        args.c2_predict = args.c2_prefix + '.predict.pb'
+
+    model = model_helper.ModelHelper(name="validation_net", init_params=False)
+
+    # Bring in the init net from init_net.pb
+    init_net_proto = caffe2_pb2.NetDef()
+    with open(args.c2_init, "rb") as f:
+        init_net_proto.ParseFromString(f.read())
+    model.param_init_net = core.Net(init_net_proto)
+
+    # bring in the predict net from predict_net.pb
+    predict_net_proto = caffe2_pb2.NetDef()
+    with open(args.c2_predict, "rb") as f:
+        predict_net_proto.ParseFromString(f.read())
+    model.net = core.Net(predict_net_proto)
+
+    data_config = resolve_data_config(None, args)
+    loader = create_loader(
+        Dataset(args.data, load_bytes=args.tf_preprocessing),
+        input_size=data_config['input_size'],
+        batch_size=args.batch_size,
+        use_prefetcher=False,
+        interpolation=data_config['interpolation'],
+        mean=data_config['mean'],
+        std=data_config['std'],
+        num_workers=args.workers,
+        crop_pct=data_config['crop_pct'],
+        tensorflow_preprocessing=args.tf_preprocessing)
+
+    # this is so obvious, wonderful interface </sarcasm>
+    input_blob = model.net.external_inputs[0]
+    output_blob = model.net.external_outputs[0]
+
+    if True:
+        device_opts = None
+    else:
+        # CUDA is crashing, no idea why, awesome error message, give it a try for kicks
+        device_opts = core.DeviceOption(caffe2_pb2.PROTO_CUDA, args.gpu_id)
+        model.net.RunAllOnGPU(gpu_id=args.gpu_id, use_cudnn=True)
+        model.param_init_net.RunAllOnGPU(gpu_id=args.gpu_id, use_cudnn=True)
+
+    model.param_init_net.GaussianFill(
+        [], input_blob.GetUnscopedName(),
+        shape=(1,) + data_config['input_size'], mean=0.0, std=1.0)
+    workspace.RunNetOnce(model.param_init_net)
+    workspace.CreateNet(model.net, overwrite=True)
+
+    batch_time = AverageMeter()
+    top1 = AverageMeter()
+    top5 = AverageMeter()
+    end = time.time()
+    for i, (input, target) in enumerate(loader):
+        # run the net and return prediction
+        caffe2_in = input.data.numpy()
+        workspace.FeedBlob(input_blob, caffe2_in, device_opts)
+        workspace.RunNet(model.net, num_iter=1)
+        output = workspace.FetchBlob(output_blob)
+
+        # measure accuracy and record loss
+        prec1, prec5 = accuracy_np(output.data, target.numpy())
+        top1.update(prec1.item(), input.size(0))
+        top5.update(prec5.item(), input.size(0))
+
+        # measure elapsed time
+        batch_time.update(time.time() - end)
+        end = time.time()
+
+        if i % args.print_freq == 0:
+            print('Test: [{0}/{1}]\t'
+                  'Time {batch_time.val:.3f} ({batch_time.avg:.3f}, {rate_avg:.3f}/s, {ms_avg:.3f} ms/sample) \t'
+                  'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
+                  'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
+                i, len(loader), batch_time=batch_time, rate_avg=input.size(0) / batch_time.avg,
+                ms_avg=100 * batch_time.avg / input.size(0), top1=top1, top5=top5))
+
+    print(' * Prec@1 {top1.avg:.3f} ({top1a:.3f}) Prec@5 {top5.avg:.3f} ({top5a:.3f})'.format(
+        top1=top1, top1a=100-top1.avg, top5=top5, top5a=100.-top5.avg))
+
+
+def accuracy_np(output, target):
+    max_indices = np.argsort(output, axis=1)[:, ::-1]
+    top5 = 100 * np.equal(max_indices[:, :5], target[:, np.newaxis]).sum(axis=1).mean()
+    top1 = 100 * np.equal(max_indices[:, 0], target).mean()
+    return top1, top5
+
+
+if __name__ == '__main__':
+    main()

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/__init__.py

@@ -0,0 +1,5 @@
+from .gen_efficientnet import *
+from .mobilenetv3 import *
+from .model_factory import create_model
+from .config import is_exportable, is_scriptable, set_exportable, set_scriptable
+from .activations import *

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/activations/__init__.py

@@ -0,0 +1,137 @@
+from geffnet import config
+from geffnet.activations.activations_me import *
+from geffnet.activations.activations_jit import *
+from geffnet.activations.activations import *
+import torch
+
+_has_silu = 'silu' in dir(torch.nn.functional)
+
+_ACT_FN_DEFAULT = dict(
+    silu=F.silu if _has_silu else swish,
+    swish=F.silu if _has_silu else swish,
+    mish=mish,
+    relu=F.relu,
+    relu6=F.relu6,
+    sigmoid=sigmoid,
+    tanh=tanh,
+    hard_sigmoid=hard_sigmoid,
+    hard_swish=hard_swish,
+)
+
+_ACT_FN_JIT = dict(
+    silu=F.silu if _has_silu else swish_jit,
+    swish=F.silu if _has_silu else swish_jit,
+    mish=mish_jit,
+)
+
+_ACT_FN_ME = dict(
+    silu=F.silu if _has_silu else swish_me,
+    swish=F.silu if _has_silu else swish_me,
+    mish=mish_me,
+    hard_swish=hard_swish_me,
+    hard_sigmoid_jit=hard_sigmoid_me,
+)
+
+_ACT_LAYER_DEFAULT = dict(
+    silu=nn.SiLU if _has_silu else Swish,
+    swish=nn.SiLU if _has_silu else Swish,
+    mish=Mish,
+    relu=nn.ReLU,
+    relu6=nn.ReLU6,
+    sigmoid=Sigmoid,
+    tanh=Tanh,
+    hard_sigmoid=HardSigmoid,
+    hard_swish=HardSwish,
+)
+
+_ACT_LAYER_JIT = dict(
+    silu=nn.SiLU if _has_silu else SwishJit,
+    swish=nn.SiLU if _has_silu else SwishJit,
+    mish=MishJit,
+)
+
+_ACT_LAYER_ME = dict(
+    silu=nn.SiLU if _has_silu else SwishMe,
+    swish=nn.SiLU if _has_silu else SwishMe,
+    mish=MishMe,
+    hard_swish=HardSwishMe,
+    hard_sigmoid=HardSigmoidMe
+)
+
+_OVERRIDE_FN = dict()
+_OVERRIDE_LAYER = dict()
+
+
+def add_override_act_fn(name, fn):
+    global _OVERRIDE_FN
+    _OVERRIDE_FN[name] = fn
+
+
+def update_override_act_fn(overrides):
+    assert isinstance(overrides, dict)
+    global _OVERRIDE_FN
+    _OVERRIDE_FN.update(overrides)
+
+
+def clear_override_act_fn():
+    global _OVERRIDE_FN
+    _OVERRIDE_FN = dict()
+
+
+def add_override_act_layer(name, fn):
+    _OVERRIDE_LAYER[name] = fn
+
+
+def update_override_act_layer(overrides):
+    assert isinstance(overrides, dict)
+    global _OVERRIDE_LAYER
+    _OVERRIDE_LAYER.update(overrides)
+
+
+def clear_override_act_layer():
+    global _OVERRIDE_LAYER
+    _OVERRIDE_LAYER = dict()
+
+
+def get_act_fn(name='relu'):
+    """ Activation Function Factory
+    Fetching activation fns by name with this function allows export or torch script friendly
+    functions to be returned dynamically based on current config.
+    """
+    if name in _OVERRIDE_FN:
+        return _OVERRIDE_FN[name]
+    use_me = not (config.is_exportable() or config.is_scriptable() or config.is_no_jit())
+    if use_me and name in _ACT_FN_ME:
+        # If not exporting or scripting the model, first look for a memory optimized version
+        # activation with custom autograd, then fallback to jit scripted, then a Python or Torch builtin
+        return _ACT_FN_ME[name]
+    if config.is_exportable() and name in ('silu', 'swish'):
+        # FIXME PyTorch SiLU doesn't ONNX export, this is a temp hack
+        return swish
+    use_jit = not (config.is_exportable() or config.is_no_jit())
+    # NOTE: export tracing should work with jit scripted components, but I keep running into issues
+    if use_jit and name in _ACT_FN_JIT:  # jit scripted models should be okay for export/scripting
+        return _ACT_FN_JIT[name]
+    return _ACT_FN_DEFAULT[name]
+
+
+def get_act_layer(name='relu'):
+    """ Activation Layer Factory
+    Fetching activation layers by name with this function allows export or torch script friendly
+    functions to be returned dynamically based on current config.
+    """
+    if name in _OVERRIDE_LAYER:
+        return _OVERRIDE_LAYER[name]
+    use_me = not (config.is_exportable() or config.is_scriptable() or config.is_no_jit())
+    if use_me and name in _ACT_LAYER_ME:
+        return _ACT_LAYER_ME[name]
+    if config.is_exportable() and name in ('silu', 'swish'):
+        # FIXME PyTorch SiLU doesn't ONNX export, this is a temp hack
+        return Swish
+    use_jit = not (config.is_exportable() or config.is_no_jit())
+    # NOTE: export tracing should work with jit scripted components, but I keep running into issues
+    if use_jit and name in _ACT_FN_JIT:  # jit scripted models should be okay for export/scripting
+        return _ACT_LAYER_JIT[name]
+    return _ACT_LAYER_DEFAULT[name]
+
+

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/activations/activations.py

@@ -0,0 +1,102 @@
+""" Activations
+
+A collection of activations fn and modules with a common interface so that they can
+easily be swapped. All have an `inplace` arg even if not used.
+
+Copyright 2020 Ross Wightman
+"""
+from torch import nn as nn
+from torch.nn import functional as F
+
+
+def swish(x, inplace: bool = False):
+    """Swish - Described originally as SiLU (https://arxiv.org/abs/1702.03118v3)
+    and also as Swish (https://arxiv.org/abs/1710.05941).
+
+    TODO Rename to SiLU with addition to PyTorch
+    """
+    return x.mul_(x.sigmoid()) if inplace else x.mul(x.sigmoid())
+
+
+class Swish(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(Swish, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        return swish(x, self.inplace)
+
+
+def mish(x, inplace: bool = False):
+    """Mish: A Self Regularized Non-Monotonic Neural Activation Function - https://arxiv.org/abs/1908.08681
+    """
+    return x.mul(F.softplus(x).tanh())
+
+
+class Mish(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(Mish, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        return mish(x, self.inplace)
+
+
+def sigmoid(x, inplace: bool = False):
+    return x.sigmoid_() if inplace else x.sigmoid()
+
+
+# PyTorch has this, but not with a consistent inplace argmument interface
+class Sigmoid(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(Sigmoid, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        return x.sigmoid_() if self.inplace else x.sigmoid()
+
+
+def tanh(x, inplace: bool = False):
+    return x.tanh_() if inplace else x.tanh()
+
+
+# PyTorch has this, but not with a consistent inplace argmument interface
+class Tanh(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(Tanh, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        return x.tanh_() if self.inplace else x.tanh()
+
+
+def hard_swish(x, inplace: bool = False):
+    inner = F.relu6(x + 3.).div_(6.)
+    return x.mul_(inner) if inplace else x.mul(inner)
+
+
+class HardSwish(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(HardSwish, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        return hard_swish(x, self.inplace)
+
+
+def hard_sigmoid(x, inplace: bool = False):
+    if inplace:
+        return x.add_(3.).clamp_(0., 6.).div_(6.)
+    else:
+        return F.relu6(x + 3.) / 6.
+
+
+class HardSigmoid(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(HardSigmoid, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        return hard_sigmoid(x, self.inplace)
+
+

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/activations/activations_jit.py

@@ -0,0 +1,79 @@
+""" Activations (jit)
+
+A collection of jit-scripted activations fn and modules with a common interface so that they can
+easily be swapped. All have an `inplace` arg even if not used.
+
+All jit scripted activations are lacking in-place variations on purpose, scripted kernel fusion does not
+currently work across in-place op boundaries, thus performance is equal to or less than the non-scripted
+versions if they contain in-place ops.
+
+Copyright 2020 Ross Wightman
+"""
+
+import torch
+from torch import nn as nn
+from torch.nn import functional as F
+
+__all__ = ['swish_jit', 'SwishJit', 'mish_jit', 'MishJit',
+           'hard_sigmoid_jit', 'HardSigmoidJit', 'hard_swish_jit', 'HardSwishJit']
+
+
+@torch.jit.script
+def swish_jit(x, inplace: bool = False):
+    """Swish - Described originally as SiLU (https://arxiv.org/abs/1702.03118v3)
+    and also as Swish (https://arxiv.org/abs/1710.05941).
+
+    TODO Rename to SiLU with addition to PyTorch
+    """
+    return x.mul(x.sigmoid())
+
+
+@torch.jit.script
+def mish_jit(x, _inplace: bool = False):
+    """Mish: A Self Regularized Non-Monotonic Neural Activation Function - https://arxiv.org/abs/1908.08681
+    """
+    return x.mul(F.softplus(x).tanh())
+
+
+class SwishJit(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(SwishJit, self).__init__()
+
+    def forward(self, x):
+        return swish_jit(x)
+
+
+class MishJit(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(MishJit, self).__init__()
+
+    def forward(self, x):
+        return mish_jit(x)
+
+
+@torch.jit.script
+def hard_sigmoid_jit(x, inplace: bool = False):
+    # return F.relu6(x + 3.) / 6.
+    return (x + 3).clamp(min=0, max=6).div(6.)  # clamp seems ever so slightly faster?
+
+
+class HardSigmoidJit(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(HardSigmoidJit, self).__init__()
+
+    def forward(self, x):
+        return hard_sigmoid_jit(x)
+
+
+@torch.jit.script
+def hard_swish_jit(x, inplace: bool = False):
+    # return x * (F.relu6(x + 3.) / 6)
+    return x * (x + 3).clamp(min=0, max=6).div(6.)  # clamp seems ever so slightly faster?
+
+
+class HardSwishJit(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(HardSwishJit, self).__init__()
+
+    def forward(self, x):
+        return hard_swish_jit(x)

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/activations/activations_me.py

@@ -0,0 +1,174 @@
+""" Activations (memory-efficient w/ custom autograd)
+
+A collection of activations fn and modules with a common interface so that they can
+easily be swapped. All have an `inplace` arg even if not used.
+
+These activations are not compatible with jit scripting or ONNX export of the model, please use either
+the JIT or basic versions of the activations.
+
+Copyright 2020 Ross Wightman
+"""
+
+import torch
+from torch import nn as nn
+from torch.nn import functional as F
+
+
+__all__ = ['swish_me', 'SwishMe', 'mish_me', 'MishMe',
+           'hard_sigmoid_me', 'HardSigmoidMe', 'hard_swish_me', 'HardSwishMe']
+
+
+@torch.jit.script
+def swish_jit_fwd(x):
+    return x.mul(torch.sigmoid(x))
+
+
+@torch.jit.script
+def swish_jit_bwd(x, grad_output):
+    x_sigmoid = torch.sigmoid(x)
+    return grad_output * (x_sigmoid * (1 + x * (1 - x_sigmoid)))
+
+
+class SwishJitAutoFn(torch.autograd.Function):
+    """ torch.jit.script optimised Swish w/ memory-efficient checkpoint
+    Inspired by conversation btw Jeremy Howard & Adam Pazske
+    https://twitter.com/jeremyphoward/status/1188251041835315200
+
+    Swish - Described originally as SiLU (https://arxiv.org/abs/1702.03118v3)
+    and also as Swish (https://arxiv.org/abs/1710.05941).
+
+    TODO Rename to SiLU with addition to PyTorch
+    """
+
+    @staticmethod
+    def forward(ctx, x):
+        ctx.save_for_backward(x)
+        return swish_jit_fwd(x)
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        x = ctx.saved_tensors[0]
+        return swish_jit_bwd(x, grad_output)
+
+
+def swish_me(x, inplace=False):
+    return SwishJitAutoFn.apply(x)
+
+
+class SwishMe(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(SwishMe, self).__init__()
+
+    def forward(self, x):
+        return SwishJitAutoFn.apply(x)
+
+
+@torch.jit.script
+def mish_jit_fwd(x):
+    return x.mul(torch.tanh(F.softplus(x)))
+
+
+@torch.jit.script
+def mish_jit_bwd(x, grad_output):
+    x_sigmoid = torch.sigmoid(x)
+    x_tanh_sp = F.softplus(x).tanh()
+    return grad_output.mul(x_tanh_sp + x * x_sigmoid * (1 - x_tanh_sp * x_tanh_sp))
+
+
+class MishJitAutoFn(torch.autograd.Function):
+    """ Mish: A Self Regularized Non-Monotonic Neural Activation Function - https://arxiv.org/abs/1908.08681
+    A memory efficient, jit scripted variant of Mish
+    """
+    @staticmethod
+    def forward(ctx, x):
+        ctx.save_for_backward(x)
+        return mish_jit_fwd(x)
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        x = ctx.saved_tensors[0]
+        return mish_jit_bwd(x, grad_output)
+
+
+def mish_me(x, inplace=False):
+    return MishJitAutoFn.apply(x)
+
+
+class MishMe(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(MishMe, self).__init__()
+
+    def forward(self, x):
+        return MishJitAutoFn.apply(x)
+
+
+@torch.jit.script
+def hard_sigmoid_jit_fwd(x, inplace: bool = False):
+    return (x + 3).clamp(min=0, max=6).div(6.)
+
+
+@torch.jit.script
+def hard_sigmoid_jit_bwd(x, grad_output):
+    m = torch.ones_like(x) * ((x >= -3.) & (x <= 3.)) / 6.
+    return grad_output * m
+
+
+class HardSigmoidJitAutoFn(torch.autograd.Function):
+    @staticmethod
+    def forward(ctx, x):
+        ctx.save_for_backward(x)
+        return hard_sigmoid_jit_fwd(x)
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        x = ctx.saved_tensors[0]
+        return hard_sigmoid_jit_bwd(x, grad_output)
+
+
+def hard_sigmoid_me(x, inplace: bool = False):
+    return HardSigmoidJitAutoFn.apply(x)
+
+
+class HardSigmoidMe(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(HardSigmoidMe, self).__init__()
+
+    def forward(self, x):
+        return HardSigmoidJitAutoFn.apply(x)
+
+
+@torch.jit.script
+def hard_swish_jit_fwd(x):
+    return x * (x + 3).clamp(min=0, max=6).div(6.)
+
+
+@torch.jit.script
+def hard_swish_jit_bwd(x, grad_output):
+    m = torch.ones_like(x) * (x >= 3.)
+    m = torch.where((x >= -3.) & (x <= 3.),  x / 3. + .5, m)
+    return grad_output * m
+
+
+class HardSwishJitAutoFn(torch.autograd.Function):
+    """A memory efficient, jit-scripted HardSwish activation"""
+    @staticmethod
+    def forward(ctx, x):
+        ctx.save_for_backward(x)
+        return hard_swish_jit_fwd(x)
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        x = ctx.saved_tensors[0]
+        return hard_swish_jit_bwd(x, grad_output)
+
+
+def hard_swish_me(x, inplace=False):
+    return HardSwishJitAutoFn.apply(x)
+
+
+class HardSwishMe(nn.Module):
+    def __init__(self, inplace: bool = False):
+        super(HardSwishMe, self).__init__()
+
+    def forward(self, x):
+        return HardSwishJitAutoFn.apply(x)

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/config.py

@@ -0,0 +1,123 @@
+""" Global layer config state
+"""
+from typing import Any, Optional
+
+__all__ = [
+    'is_exportable', 'is_scriptable', 'is_no_jit', 'layer_config_kwargs',
+    'set_exportable', 'set_scriptable', 'set_no_jit', 'set_layer_config'
+]
+
+# Set to True if prefer to have layers with no jit optimization (includes activations)
+_NO_JIT = False
+
+# Set to True if prefer to have activation layers with no jit optimization
+# NOTE not currently used as no difference between no_jit and no_activation jit as only layers obeying
+# the jit flags so far are activations. This will change as more layers are updated and/or added.
+_NO_ACTIVATION_JIT = False
+
+# Set to True if exporting a model with Same padding via ONNX
+_EXPORTABLE = False
+
+# Set to True if wanting to use torch.jit.script on a model
+_SCRIPTABLE = False
+
+
+def is_no_jit():
+    return _NO_JIT
+
+
+class set_no_jit:
+    def __init__(self, mode: bool) -> None:
+        global _NO_JIT
+        self.prev = _NO_JIT
+        _NO_JIT = mode
+
+    def __enter__(self) -> None:
+        pass
+
+    def __exit__(self, *args: Any) -> bool:
+        global _NO_JIT
+        _NO_JIT = self.prev
+        return False
+
+
+def is_exportable():
+    return _EXPORTABLE
+
+
+class set_exportable:
+    def __init__(self, mode: bool) -> None:
+        global _EXPORTABLE
+        self.prev = _EXPORTABLE
+        _EXPORTABLE = mode
+
+    def __enter__(self) -> None:
+        pass
+
+    def __exit__(self, *args: Any) -> bool:
+        global _EXPORTABLE
+        _EXPORTABLE = self.prev
+        return False
+
+
+def is_scriptable():
+    return _SCRIPTABLE
+
+
+class set_scriptable:
+    def __init__(self, mode: bool) -> None:
+        global _SCRIPTABLE
+        self.prev = _SCRIPTABLE
+        _SCRIPTABLE = mode
+
+    def __enter__(self) -> None:
+        pass
+
+    def __exit__(self, *args: Any) -> bool:
+        global _SCRIPTABLE
+        _SCRIPTABLE = self.prev
+        return False
+
+
+class set_layer_config:
+    """ Layer config context manager that allows setting all layer config flags at once.
+    If a flag arg is None, it will not change the current value.
+    """
+    def __init__(
+            self,
+            scriptable: Optional[bool] = None,
+            exportable: Optional[bool] = None,
+            no_jit: Optional[bool] = None,
+            no_activation_jit: Optional[bool] = None):
+        global _SCRIPTABLE
+        global _EXPORTABLE
+        global _NO_JIT
+        global _NO_ACTIVATION_JIT
+        self.prev = _SCRIPTABLE, _EXPORTABLE, _NO_JIT, _NO_ACTIVATION_JIT
+        if scriptable is not None:
+            _SCRIPTABLE = scriptable
+        if exportable is not None:
+            _EXPORTABLE = exportable
+        if no_jit is not None:
+            _NO_JIT = no_jit
+        if no_activation_jit is not None:
+            _NO_ACTIVATION_JIT = no_activation_jit
+
+    def __enter__(self) -> None:
+        pass
+
+    def __exit__(self, *args: Any) -> bool:
+        global _SCRIPTABLE
+        global _EXPORTABLE
+        global _NO_JIT
+        global _NO_ACTIVATION_JIT
+        _SCRIPTABLE, _EXPORTABLE, _NO_JIT, _NO_ACTIVATION_JIT = self.prev
+        return False
+
+
+def layer_config_kwargs(kwargs):
+    """ Consume config kwargs and return contextmgr obj """
+    return set_layer_config(
+        scriptable=kwargs.pop('scriptable', None),
+        exportable=kwargs.pop('exportable', None),
+        no_jit=kwargs.pop('no_jit', None))

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/conv2d_layers.py

@@ -0,0 +1,304 @@
+""" Conv2D w/ SAME padding, CondConv, MixedConv
+
+A collection of conv layers and padding helpers needed by EfficientNet, MixNet, and
+MobileNetV3 models that maintain weight compatibility with original Tensorflow models.
+
+Copyright 2020 Ross Wightman
+"""
+import collections.abc
+import math
+from functools import partial
+from itertools import repeat
+from typing import Tuple, Optional
+
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .config import *
+
+
+# From PyTorch internals
+def _ntuple(n):
+    def parse(x):
+        if isinstance(x, collections.abc.Iterable):
+            return x
+        return tuple(repeat(x, n))
+    return parse
+
+
+_single = _ntuple(1)
+_pair = _ntuple(2)
+_triple = _ntuple(3)
+_quadruple = _ntuple(4)
+
+
+def _is_static_pad(kernel_size, stride=1, dilation=1, **_):
+    return stride == 1 and (dilation * (kernel_size - 1)) % 2 == 0
+
+
+def _get_padding(kernel_size, stride=1, dilation=1, **_):
+    padding = ((stride - 1) + dilation * (kernel_size - 1)) // 2
+    return padding
+
+
+def _calc_same_pad(i: int, k: int, s: int, d: int):
+    return max((-(i // -s) - 1) * s + (k - 1) * d + 1 - i, 0)
+
+
+def _same_pad_arg(input_size, kernel_size, stride, dilation):
+    ih, iw = input_size
+    kh, kw = kernel_size
+    pad_h = _calc_same_pad(ih, kh, stride[0], dilation[0])
+    pad_w = _calc_same_pad(iw, kw, stride[1], dilation[1])
+    return [pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2]
+
+
+def _split_channels(num_chan, num_groups):
+    split = [num_chan // num_groups for _ in range(num_groups)]
+    split[0] += num_chan - sum(split)
+    return split
+
+
+def conv2d_same(
+        x, weight: torch.Tensor, bias: Optional[torch.Tensor] = None, stride: Tuple[int, int] = (1, 1),
+        padding: Tuple[int, int] = (0, 0), dilation: Tuple[int, int] = (1, 1), groups: int = 1):
+    ih, iw = x.size()[-2:]
+    kh, kw = weight.size()[-2:]
+    pad_h = _calc_same_pad(ih, kh, stride[0], dilation[0])
+    pad_w = _calc_same_pad(iw, kw, stride[1], dilation[1])
+    x = F.pad(x, [pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2])
+    return F.conv2d(x, weight, bias, stride, (0, 0), dilation, groups)
+
+
+class Conv2dSame(nn.Conv2d):
+    """ Tensorflow like 'SAME' convolution wrapper for 2D convolutions
+    """
+
+    # pylint: disable=unused-argument
+    def __init__(self, in_channels, out_channels, kernel_size, stride=1,
+                 padding=0, dilation=1, groups=1, bias=True):
+        super(Conv2dSame, self).__init__(
+            in_channels, out_channels, kernel_size, stride, 0, dilation, groups, bias)
+
+    def forward(self, x):
+        return conv2d_same(x, self.weight, self.bias, self.stride, self.padding, self.dilation, self.groups)
+
+
+class Conv2dSameExport(nn.Conv2d):
+    """ ONNX export friendly Tensorflow like 'SAME' convolution wrapper for 2D convolutions
+
+    NOTE: This does not currently work with torch.jit.script
+    """
+
+    # pylint: disable=unused-argument
+    def __init__(self, in_channels, out_channels, kernel_size, stride=1,
+                 padding=0, dilation=1, groups=1, bias=True):
+        super(Conv2dSameExport, self).__init__(
+            in_channels, out_channels, kernel_size, stride, 0, dilation, groups, bias)
+        self.pad = None
+        self.pad_input_size = (0, 0)
+
+    def forward(self, x):
+        input_size = x.size()[-2:]
+        if self.pad is None:
+            pad_arg = _same_pad_arg(input_size, self.weight.size()[-2:], self.stride, self.dilation)
+            self.pad = nn.ZeroPad2d(pad_arg)
+            self.pad_input_size = input_size
+
+        if self.pad is not None:
+            x = self.pad(x)
+        return F.conv2d(
+            x, self.weight, self.bias, self.stride, self.padding, self.dilation, self.groups)
+
+
+def get_padding_value(padding, kernel_size, **kwargs):
+    dynamic = False
+    if isinstance(padding, str):
+        # for any string padding, the padding will be calculated for you, one of three ways
+        padding = padding.lower()
+        if padding == 'same':
+            # TF compatible 'SAME' padding, has a performance and GPU memory allocation impact
+            if _is_static_pad(kernel_size, **kwargs):
+                # static case, no extra overhead
+                padding = _get_padding(kernel_size, **kwargs)
+            else:
+                # dynamic padding
+                padding = 0
+                dynamic = True
+        elif padding == 'valid':
+            # 'VALID' padding, same as padding=0
+            padding = 0
+        else:
+            # Default to PyTorch style 'same'-ish symmetric padding
+            padding = _get_padding(kernel_size, **kwargs)
+    return padding, dynamic
+
+
+def create_conv2d_pad(in_chs, out_chs, kernel_size, **kwargs):
+    padding = kwargs.pop('padding', '')
+    kwargs.setdefault('bias', False)
+    padding, is_dynamic = get_padding_value(padding, kernel_size, **kwargs)
+    if is_dynamic:
+        if is_exportable():
+            assert not is_scriptable()
+            return Conv2dSameExport(in_chs, out_chs, kernel_size, **kwargs)
+        else:
+            return Conv2dSame(in_chs, out_chs, kernel_size, **kwargs)
+    else:
+        return nn.Conv2d(in_chs, out_chs, kernel_size, padding=padding, **kwargs)
+
+
+class MixedConv2d(nn.ModuleDict):
+    """ Mixed Grouped Convolution
+    Based on MDConv and GroupedConv in MixNet impl:
+      https://github.com/tensorflow/tpu/blob/master/models/official/mnasnet/mixnet/custom_layers.py
+    """
+
+    def __init__(self, in_channels, out_channels, kernel_size=3,
+                 stride=1, padding='', dilation=1, depthwise=False, **kwargs):
+        super(MixedConv2d, self).__init__()
+
+        kernel_size = kernel_size if isinstance(kernel_size, list) else [kernel_size]
+        num_groups = len(kernel_size)
+        in_splits = _split_channels(in_channels, num_groups)
+        out_splits = _split_channels(out_channels, num_groups)
+        self.in_channels = sum(in_splits)
+        self.out_channels = sum(out_splits)
+        for idx, (k, in_ch, out_ch) in enumerate(zip(kernel_size, in_splits, out_splits)):
+            conv_groups = out_ch if depthwise else 1
+            self.add_module(
+                str(idx),
+                create_conv2d_pad(
+                    in_ch, out_ch, k, stride=stride,
+                    padding=padding, dilation=dilation, groups=conv_groups, **kwargs)
+            )
+        self.splits = in_splits
+
+    def forward(self, x):
+        x_split = torch.split(x, self.splits, 1)
+        x_out = [conv(x_split[i]) for i, conv in enumerate(self.values())]
+        x = torch.cat(x_out, 1)
+        return x
+
+
+def get_condconv_initializer(initializer, num_experts, expert_shape):
+    def condconv_initializer(weight):
+        """CondConv initializer function."""
+        num_params = np.prod(expert_shape)
+        if (len(weight.shape) != 2 or weight.shape[0] != num_experts or
+                weight.shape[1] != num_params):
+            raise (ValueError(
+                'CondConv variables must have shape [num_experts, num_params]'))
+        for i in range(num_experts):
+            initializer(weight[i].view(expert_shape))
+    return condconv_initializer
+
+
+class CondConv2d(nn.Module):
+    """ Conditional Convolution
+    Inspired by: https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/condconv/condconv_layers.py
+
+    Grouped convolution hackery for parallel execution of the per-sample kernel filters inspired by this discussion:
+    https://github.com/pytorch/pytorch/issues/17983
+    """
+    __constants__ = ['bias', 'in_channels', 'out_channels', 'dynamic_padding']
+
+    def __init__(self, in_channels, out_channels, kernel_size=3,
+                 stride=1, padding='', dilation=1, groups=1, bias=False, num_experts=4):
+        super(CondConv2d, self).__init__()
+
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.kernel_size = _pair(kernel_size)
+        self.stride = _pair(stride)
+        padding_val, is_padding_dynamic = get_padding_value(
+            padding, kernel_size, stride=stride, dilation=dilation)
+        self.dynamic_padding = is_padding_dynamic  # if in forward to work with torchscript
+        self.padding = _pair(padding_val)
+        self.dilation = _pair(dilation)
+        self.groups = groups
+        self.num_experts = num_experts
+
+        self.weight_shape = (self.out_channels, self.in_channels // self.groups) + self.kernel_size
+        weight_num_param = 1
+        for wd in self.weight_shape:
+            weight_num_param *= wd
+        self.weight = torch.nn.Parameter(torch.Tensor(self.num_experts, weight_num_param))
+
+        if bias:
+            self.bias_shape = (self.out_channels,)
+            self.bias = torch.nn.Parameter(torch.Tensor(self.num_experts, self.out_channels))
+        else:
+            self.register_parameter('bias', None)
+
+        self.reset_parameters()
+
+    def reset_parameters(self):
+        init_weight = get_condconv_initializer(
+            partial(nn.init.kaiming_uniform_, a=math.sqrt(5)), self.num_experts, self.weight_shape)
+        init_weight(self.weight)
+        if self.bias is not None:
+            fan_in = np.prod(self.weight_shape[1:])
+            bound = 1 / math.sqrt(fan_in)
+            init_bias = get_condconv_initializer(
+                partial(nn.init.uniform_, a=-bound, b=bound), self.num_experts, self.bias_shape)
+            init_bias(self.bias)
+
+    def forward(self, x, routing_weights):
+        B, C, H, W = x.shape
+        weight = torch.matmul(routing_weights, self.weight)
+        new_weight_shape = (B * self.out_channels, self.in_channels // self.groups) + self.kernel_size
+        weight = weight.view(new_weight_shape)
+        bias = None
+        if self.bias is not None:
+            bias = torch.matmul(routing_weights, self.bias)
+            bias = bias.view(B * self.out_channels)
+        # move batch elements with channels so each batch element can be efficiently convolved with separate kernel
+        x = x.view(1, B * C, H, W)
+        if self.dynamic_padding:
+            out = conv2d_same(
+                x, weight, bias, stride=self.stride, padding=self.padding,
+                dilation=self.dilation, groups=self.groups * B)
+        else:
+            out = F.conv2d(
+                x, weight, bias, stride=self.stride, padding=self.padding,
+                dilation=self.dilation, groups=self.groups * B)
+        out = out.permute([1, 0, 2, 3]).view(B, self.out_channels, out.shape[-2], out.shape[-1])
+
+        # Literal port (from TF definition)
+        # x = torch.split(x, 1, 0)
+        # weight = torch.split(weight, 1, 0)
+        # if self.bias is not None:
+        #     bias = torch.matmul(routing_weights, self.bias)
+        #     bias = torch.split(bias, 1, 0)
+        # else:
+        #     bias = [None] * B
+        # out = []
+        # for xi, wi, bi in zip(x, weight, bias):
+        #     wi = wi.view(*self.weight_shape)
+        #     if bi is not None:
+        #         bi = bi.view(*self.bias_shape)
+        #     out.append(self.conv_fn(
+        #         xi, wi, bi, stride=self.stride, padding=self.padding,
+        #         dilation=self.dilation, groups=self.groups))
+        # out = torch.cat(out, 0)
+        return out
+
+
+def select_conv2d(in_chs, out_chs, kernel_size, **kwargs):
+    assert 'groups' not in kwargs  # only use 'depthwise' bool arg
+    if isinstance(kernel_size, list):
+        assert 'num_experts' not in kwargs  # MixNet + CondConv combo not supported currently
+        # We're going to use only lists for defining the MixedConv2d kernel groups,
+        # ints, tuples, other iterables will continue to pass to normal conv and specify h, w.
+        m = MixedConv2d(in_chs, out_chs, kernel_size, **kwargs)
+    else:
+        depthwise = kwargs.pop('depthwise', False)
+        groups = out_chs if depthwise else 1
+        if 'num_experts' in kwargs and kwargs['num_experts'] > 0:
+            m = CondConv2d(in_chs, out_chs, kernel_size, groups=groups, **kwargs)
+        else:
+            m = create_conv2d_pad(in_chs, out_chs, kernel_size, groups=groups, **kwargs)
+    return m

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/efficientnet_builder.py

@@ -0,0 +1,683 @@
+""" EfficientNet / MobileNetV3 Blocks and Builder
+
+Copyright 2020 Ross Wightman
+"""
+import re
+from copy import deepcopy
+
+from .conv2d_layers import *
+from geffnet.activations import *
+
+__all__ = ['get_bn_args_tf', 'resolve_bn_args', 'resolve_se_args', 'resolve_act_layer', 'make_divisible',
+           'round_channels', 'drop_connect', 'SqueezeExcite', 'ConvBnAct', 'DepthwiseSeparableConv',
+           'InvertedResidual', 'CondConvResidual', 'EdgeResidual', 'EfficientNetBuilder', 'decode_arch_def',
+           'initialize_weight_default', 'initialize_weight_goog', 'BN_MOMENTUM_TF_DEFAULT', 'BN_EPS_TF_DEFAULT'
+]
+
+# Defaults used for Google/Tensorflow training of mobile networks /w RMSprop as per
+# papers and TF reference implementations. PT momentum equiv for TF decay is (1 - TF decay)
+# NOTE: momentum varies btw .99 and .9997 depending on source
+# .99 in official TF TPU impl
+# .9997 (/w .999 in search space) for paper
+#
+# PyTorch defaults are momentum = .1, eps = 1e-5
+#
+BN_MOMENTUM_TF_DEFAULT = 1 - 0.99
+BN_EPS_TF_DEFAULT = 1e-3
+_BN_ARGS_TF = dict(momentum=BN_MOMENTUM_TF_DEFAULT, eps=BN_EPS_TF_DEFAULT)
+
+
+def get_bn_args_tf():
+    return _BN_ARGS_TF.copy()
+
+
+def resolve_bn_args(kwargs):
+    bn_args = get_bn_args_tf() if kwargs.pop('bn_tf', False) else {}
+    bn_momentum = kwargs.pop('bn_momentum', None)
+    if bn_momentum is not None:
+        bn_args['momentum'] = bn_momentum
+    bn_eps = kwargs.pop('bn_eps', None)
+    if bn_eps is not None:
+        bn_args['eps'] = bn_eps
+    return bn_args
+
+
+_SE_ARGS_DEFAULT = dict(
+    gate_fn=sigmoid,
+    act_layer=None,  # None == use containing block's activation layer
+    reduce_mid=False,
+    divisor=1)
+
+
+def resolve_se_args(kwargs, in_chs, act_layer=None):
+    se_kwargs = kwargs.copy() if kwargs is not None else {}
+    # fill in args that aren't specified with the defaults
+    for k, v in _SE_ARGS_DEFAULT.items():
+        se_kwargs.setdefault(k, v)
+    # some models, like MobilNetV3, calculate SE reduction chs from the containing block's mid_ch instead of in_ch
+    if not se_kwargs.pop('reduce_mid'):
+        se_kwargs['reduced_base_chs'] = in_chs
+    # act_layer override, if it remains None, the containing block's act_layer will be used
+    if se_kwargs['act_layer'] is None:
+        assert act_layer is not None
+        se_kwargs['act_layer'] = act_layer
+    return se_kwargs
+
+
+def resolve_act_layer(kwargs, default='relu'):
+    act_layer = kwargs.pop('act_layer', default)
+    if isinstance(act_layer, str):
+        act_layer = get_act_layer(act_layer)
+    return act_layer
+
+
+def make_divisible(v: int, divisor: int = 8, min_value: int = None):
+    min_value = min_value or divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    if new_v < 0.9 * v:  # ensure round down does not go down by more than 10%.
+        new_v += divisor
+    return new_v
+
+
+def round_channels(channels, multiplier=1.0, divisor=8, channel_min=None):
+    """Round number of filters based on depth multiplier."""
+    if not multiplier:
+        return channels
+    channels *= multiplier
+    return make_divisible(channels, divisor, channel_min)
+
+
+def drop_connect(inputs, training: bool = False, drop_connect_rate: float = 0.):
+    """Apply drop connect."""
+    if not training:
+        return inputs
+
+    keep_prob = 1 - drop_connect_rate
+    random_tensor = keep_prob + torch.rand(
+        (inputs.size()[0], 1, 1, 1), dtype=inputs.dtype, device=inputs.device)
+    random_tensor.floor_()  # binarize
+    output = inputs.div(keep_prob) * random_tensor
+    return output
+
+
+class SqueezeExcite(nn.Module):
+
+    def __init__(self, in_chs, se_ratio=0.25, reduced_base_chs=None, act_layer=nn.ReLU, gate_fn=sigmoid, divisor=1):
+        super(SqueezeExcite, self).__init__()
+        reduced_chs = make_divisible((reduced_base_chs or in_chs) * se_ratio, divisor)
+        self.conv_reduce = nn.Conv2d(in_chs, reduced_chs, 1, bias=True)
+        self.act1 = act_layer(inplace=True)
+        self.conv_expand = nn.Conv2d(reduced_chs, in_chs, 1, bias=True)
+        self.gate_fn = gate_fn
+
+    def forward(self, x):
+        x_se = x.mean((2, 3), keepdim=True)
+        x_se = self.conv_reduce(x_se)
+        x_se = self.act1(x_se)
+        x_se = self.conv_expand(x_se)
+        x = x * self.gate_fn(x_se)
+        return x
+
+
+class ConvBnAct(nn.Module):
+    def __init__(self, in_chs, out_chs, kernel_size,
+                 stride=1, pad_type='', act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d, norm_kwargs=None):
+        super(ConvBnAct, self).__init__()
+        assert stride in [1, 2]
+        norm_kwargs = norm_kwargs or {}
+        self.conv = select_conv2d(in_chs, out_chs, kernel_size, stride=stride, padding=pad_type)
+        self.bn1 = norm_layer(out_chs, **norm_kwargs)
+        self.act1 = act_layer(inplace=True)
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn1(x)
+        x = self.act1(x)
+        return x
+
+
+class DepthwiseSeparableConv(nn.Module):
+    """ DepthwiseSeparable block
+    Used for DS convs in MobileNet-V1 and in the place of IR blocks with an expansion
+    factor of 1.0. This is an alternative to having a IR with optional first pw conv.
+    """
+    def __init__(self, in_chs, out_chs, dw_kernel_size=3,
+                 stride=1, pad_type='', act_layer=nn.ReLU, noskip=False,
+                 pw_kernel_size=1, pw_act=False, se_ratio=0., se_kwargs=None,
+                 norm_layer=nn.BatchNorm2d, norm_kwargs=None, drop_connect_rate=0.):
+        super(DepthwiseSeparableConv, self).__init__()
+        assert stride in [1, 2]
+        norm_kwargs = norm_kwargs or {}
+        self.has_residual = (stride == 1 and in_chs == out_chs) and not noskip
+        self.drop_connect_rate = drop_connect_rate
+
+        self.conv_dw = select_conv2d(
+            in_chs, in_chs, dw_kernel_size, stride=stride, padding=pad_type, depthwise=True)
+        self.bn1 = norm_layer(in_chs, **norm_kwargs)
+        self.act1 = act_layer(inplace=True)
+
+        # Squeeze-and-excitation
+        if se_ratio is not None and se_ratio > 0.:
+            se_kwargs = resolve_se_args(se_kwargs, in_chs, act_layer)
+            self.se = SqueezeExcite(in_chs, se_ratio=se_ratio, **se_kwargs)
+        else:
+            self.se = nn.Identity()
+
+        self.conv_pw = select_conv2d(in_chs, out_chs, pw_kernel_size, padding=pad_type)
+        self.bn2 = norm_layer(out_chs, **norm_kwargs)
+        self.act2 = act_layer(inplace=True) if pw_act else nn.Identity()
+
+    def forward(self, x):
+        residual = x
+
+        x = self.conv_dw(x)
+        x = self.bn1(x)
+        x = self.act1(x)
+
+        x = self.se(x)
+
+        x = self.conv_pw(x)
+        x = self.bn2(x)
+        x = self.act2(x)
+
+        if self.has_residual:
+            if self.drop_connect_rate > 0.:
+                x = drop_connect(x, self.training, self.drop_connect_rate)
+            x += residual
+        return x
+
+
+class InvertedResidual(nn.Module):
+    """ Inverted residual block w/ optional SE"""
+
+    def __init__(self, in_chs, out_chs, dw_kernel_size=3,
+                 stride=1, pad_type='', act_layer=nn.ReLU, noskip=False,
+                 exp_ratio=1.0, exp_kernel_size=1, pw_kernel_size=1,
+                 se_ratio=0., se_kwargs=None, norm_layer=nn.BatchNorm2d, norm_kwargs=None,
+                 conv_kwargs=None, drop_connect_rate=0.):
+        super(InvertedResidual, self).__init__()
+        norm_kwargs = norm_kwargs or {}
+        conv_kwargs = conv_kwargs or {}
+        mid_chs: int = make_divisible(in_chs * exp_ratio)
+        self.has_residual = (in_chs == out_chs and stride == 1) and not noskip
+        self.drop_connect_rate = drop_connect_rate
+
+        # Point-wise expansion
+        self.conv_pw = select_conv2d(in_chs, mid_chs, exp_kernel_size, padding=pad_type, **conv_kwargs)
+        self.bn1 = norm_layer(mid_chs, **norm_kwargs)
+        self.act1 = act_layer(inplace=True)
+
+        # Depth-wise convolution
+        self.conv_dw = select_conv2d(
+            mid_chs, mid_chs, dw_kernel_size, stride=stride, padding=pad_type, depthwise=True, **conv_kwargs)
+        self.bn2 = norm_layer(mid_chs, **norm_kwargs)
+        self.act2 = act_layer(inplace=True)
+
+        # Squeeze-and-excitation
+        if se_ratio is not None and se_ratio > 0.:
+            se_kwargs = resolve_se_args(se_kwargs, in_chs, act_layer)
+            self.se = SqueezeExcite(mid_chs, se_ratio=se_ratio, **se_kwargs)
+        else:
+            self.se = nn.Identity()  # for jit.script compat
+
+        # Point-wise linear projection
+        self.conv_pwl = select_conv2d(mid_chs, out_chs, pw_kernel_size, padding=pad_type, **conv_kwargs)
+        self.bn3 = norm_layer(out_chs, **norm_kwargs)
+
+    def forward(self, x):
+        residual = x
+
+        # Point-wise expansion
+        x = self.conv_pw(x)
+        x = self.bn1(x)
+        x = self.act1(x)
+
+        # Depth-wise convolution
+        x = self.conv_dw(x)
+        x = self.bn2(x)
+        x = self.act2(x)
+
+        # Squeeze-and-excitation
+        x = self.se(x)
+
+        # Point-wise linear projection
+        x = self.conv_pwl(x)
+        x = self.bn3(x)
+
+        if self.has_residual:
+            if self.drop_connect_rate > 0.:
+                x = drop_connect(x, self.training, self.drop_connect_rate)
+            x += residual
+        return x
+
+
+class CondConvResidual(InvertedResidual):
+    """ Inverted residual block w/ CondConv routing"""
+
+    def __init__(self, in_chs, out_chs, dw_kernel_size=3,
+                 stride=1, pad_type='', act_layer=nn.ReLU, noskip=False,
+                 exp_ratio=1.0, exp_kernel_size=1, pw_kernel_size=1,
+                 se_ratio=0., se_kwargs=None, norm_layer=nn.BatchNorm2d, norm_kwargs=None,
+                 num_experts=0, drop_connect_rate=0.):
+
+        self.num_experts = num_experts
+        conv_kwargs = dict(num_experts=self.num_experts)
+
+        super(CondConvResidual, self).__init__(
+            in_chs, out_chs, dw_kernel_size=dw_kernel_size, stride=stride, pad_type=pad_type,
+            act_layer=act_layer, noskip=noskip, exp_ratio=exp_ratio, exp_kernel_size=exp_kernel_size,
+            pw_kernel_size=pw_kernel_size, se_ratio=se_ratio, se_kwargs=se_kwargs,
+            norm_layer=norm_layer, norm_kwargs=norm_kwargs, conv_kwargs=conv_kwargs,
+            drop_connect_rate=drop_connect_rate)
+
+        self.routing_fn = nn.Linear(in_chs, self.num_experts)
+
+    def forward(self, x):
+        residual = x
+
+        # CondConv routing
+        pooled_inputs = F.adaptive_avg_pool2d(x, 1).flatten(1)
+        routing_weights = torch.sigmoid(self.routing_fn(pooled_inputs))
+
+        # Point-wise expansion
+        x = self.conv_pw(x, routing_weights)
+        x = self.bn1(x)
+        x = self.act1(x)
+
+        # Depth-wise convolution
+        x = self.conv_dw(x, routing_weights)
+        x = self.bn2(x)
+        x = self.act2(x)
+
+        # Squeeze-and-excitation
+        x = self.se(x)
+
+        # Point-wise linear projection
+        x = self.conv_pwl(x, routing_weights)
+        x = self.bn3(x)
+
+        if self.has_residual:
+            if self.drop_connect_rate > 0.:
+                x = drop_connect(x, self.training, self.drop_connect_rate)
+            x += residual
+        return x
+
+
+class EdgeResidual(nn.Module):
+    """ EdgeTPU Residual block with expansion convolution followed by pointwise-linear w/ stride"""
+
+    def __init__(self, in_chs, out_chs, exp_kernel_size=3, exp_ratio=1.0, fake_in_chs=0,
+                 stride=1, pad_type='', act_layer=nn.ReLU, noskip=False, pw_kernel_size=1,
+                 se_ratio=0., se_kwargs=None, norm_layer=nn.BatchNorm2d, norm_kwargs=None, drop_connect_rate=0.):
+        super(EdgeResidual, self).__init__()
+        norm_kwargs = norm_kwargs or {}
+        mid_chs = make_divisible(fake_in_chs * exp_ratio) if fake_in_chs > 0 else make_divisible(in_chs * exp_ratio)
+        self.has_residual = (in_chs == out_chs and stride == 1) and not noskip
+        self.drop_connect_rate = drop_connect_rate
+
+        # Expansion convolution
+        self.conv_exp = select_conv2d(in_chs, mid_chs, exp_kernel_size, padding=pad_type)
+        self.bn1 = norm_layer(mid_chs, **norm_kwargs)
+        self.act1 = act_layer(inplace=True)
+
+        # Squeeze-and-excitation
+        if se_ratio is not None and se_ratio > 0.:
+            se_kwargs = resolve_se_args(se_kwargs, in_chs, act_layer)
+            self.se = SqueezeExcite(mid_chs, se_ratio=se_ratio, **se_kwargs)
+        else:
+            self.se = nn.Identity()
+
+        # Point-wise linear projection
+        self.conv_pwl = select_conv2d(mid_chs, out_chs, pw_kernel_size, stride=stride, padding=pad_type)
+        self.bn2 = nn.BatchNorm2d(out_chs, **norm_kwargs)
+
+    def forward(self, x):
+        residual = x
+
+        # Expansion convolution
+        x = self.conv_exp(x)
+        x = self.bn1(x)
+        x = self.act1(x)
+
+        # Squeeze-and-excitation
+        x = self.se(x)
+
+        # Point-wise linear projection
+        x = self.conv_pwl(x)
+        x = self.bn2(x)
+
+        if self.has_residual:
+            if self.drop_connect_rate > 0.:
+                x = drop_connect(x, self.training, self.drop_connect_rate)
+            x += residual
+
+        return x
+
+
+class EfficientNetBuilder:
+    """ Build Trunk Blocks for Efficient/Mobile Networks
+
+    This ended up being somewhat of a cross between
+    https://github.com/tensorflow/tpu/blob/master/models/official/mnasnet/mnasnet_models.py
+    and
+    https://github.com/facebookresearch/maskrcnn-benchmark/blob/master/maskrcnn_benchmark/modeling/backbone/fbnet_builder.py
+
+    """
+
+    def __init__(self, channel_multiplier=1.0, channel_divisor=8, channel_min=None,
+                 pad_type='', act_layer=None, se_kwargs=None,
+                 norm_layer=nn.BatchNorm2d, norm_kwargs=None, drop_connect_rate=0.):
+        self.channel_multiplier = channel_multiplier
+        self.channel_divisor = channel_divisor
+        self.channel_min = channel_min
+        self.pad_type = pad_type
+        self.act_layer = act_layer
+        self.se_kwargs = se_kwargs
+        self.norm_layer = norm_layer
+        self.norm_kwargs = norm_kwargs
+        self.drop_connect_rate = drop_connect_rate
+
+        # updated during build
+        self.in_chs = None
+        self.block_idx = 0
+        self.block_count = 0
+
+    def _round_channels(self, chs):
+        return round_channels(chs, self.channel_multiplier, self.channel_divisor, self.channel_min)
+
+    def _make_block(self, ba):
+        bt = ba.pop('block_type')
+        ba['in_chs'] = self.in_chs
+        ba['out_chs'] = self._round_channels(ba['out_chs'])
+        if 'fake_in_chs' in ba and ba['fake_in_chs']:
+            # FIXME this is a hack to work around mismatch in origin impl input filters for EdgeTPU
+            ba['fake_in_chs'] = self._round_channels(ba['fake_in_chs'])
+        ba['norm_layer'] = self.norm_layer
+        ba['norm_kwargs'] = self.norm_kwargs
+        ba['pad_type'] = self.pad_type
+        # block act fn overrides the model default
+        ba['act_layer'] = ba['act_layer'] if ba['act_layer'] is not None else self.act_layer
+        assert ba['act_layer'] is not None
+        if bt == 'ir':
+            ba['drop_connect_rate'] = self.drop_connect_rate * self.block_idx / self.block_count
+            ba['se_kwargs'] = self.se_kwargs
+            if ba.get('num_experts', 0) > 0:
+                block = CondConvResidual(**ba)
+            else:
+                block = InvertedResidual(**ba)
+        elif bt == 'ds' or bt == 'dsa':
+            ba['drop_connect_rate'] = self.drop_connect_rate * self.block_idx / self.block_count
+            ba['se_kwargs'] = self.se_kwargs
+            block = DepthwiseSeparableConv(**ba)
+        elif bt == 'er':
+            ba['drop_connect_rate'] = self.drop_connect_rate * self.block_idx / self.block_count
+            ba['se_kwargs'] = self.se_kwargs
+            block = EdgeResidual(**ba)
+        elif bt == 'cn':
+            block = ConvBnAct(**ba)
+        else:
+            assert False, 'Uknkown block type (%s) while building model.' % bt
+        self.in_chs = ba['out_chs']  # update in_chs for arg of next block
+        return block
+
+    def _make_stack(self, stack_args):
+        blocks = []
+        # each stack (stage) contains a list of block arguments
+        for i, ba in enumerate(stack_args):
+            if i >= 1:
+                # only the first block in any stack can have a stride > 1
+                ba['stride'] = 1
+            block = self._make_block(ba)
+            blocks.append(block)
+            self.block_idx += 1  # incr global idx (across all stacks)
+        return nn.Sequential(*blocks)
+
+    def __call__(self, in_chs, block_args):
+        """ Build the blocks
+        Args:
+            in_chs: Number of input-channels passed to first block
+            block_args: A list of lists, outer list defines stages, inner
+                list contains strings defining block configuration(s)
+        Return:
+             List of block stacks (each stack wrapped in nn.Sequential)
+        """
+        self.in_chs = in_chs
+        self.block_count = sum([len(x) for x in block_args])
+        self.block_idx = 0
+        blocks = []
+        # outer list of block_args defines the stacks ('stages' by some conventions)
+        for stack_idx, stack in enumerate(block_args):
+            assert isinstance(stack, list)
+            stack = self._make_stack(stack)
+            blocks.append(stack)
+        return blocks
+
+
+def _parse_ksize(ss):
+    if ss.isdigit():
+        return int(ss)
+    else:
+        return [int(k) for k in ss.split('.')]
+
+
+def _decode_block_str(block_str):
+    """ Decode block definition string
+
+    Gets a list of block arg (dicts) through a string notation of arguments.
+    E.g. ir_r2_k3_s2_e1_i32_o16_se0.25_noskip
+
+    All args can exist in any order with the exception of the leading string which
+    is assumed to indicate the block type.
+
+    leading string - block type (
+      ir = InvertedResidual, ds = DepthwiseSep, dsa = DeptwhiseSep with pw act, cn = ConvBnAct)
+    r - number of repeat blocks,
+    k - kernel size,
+    s - strides (1-9),
+    e - expansion ratio,
+    c - output channels,
+    se - squeeze/excitation ratio
+    n - activation fn ('re', 'r6', 'hs', or 'sw')
+    Args:
+        block_str: a string representation of block arguments.
+    Returns:
+        A list of block args (dicts)
+    Raises:
+        ValueError: if the string def not properly specified (TODO)
+    """
+    assert isinstance(block_str, str)
+    ops = block_str.split('_')
+    block_type = ops[0]  # take the block type off the front
+    ops = ops[1:]
+    options = {}
+    noskip = False
+    for op in ops:
+        # string options being checked on individual basis, combine if they grow
+        if op == 'noskip':
+            noskip = True
+        elif op.startswith('n'):
+            # activation fn
+            key = op[0]
+            v = op[1:]
+            if v == 're':
+                value = get_act_layer('relu')
+            elif v == 'r6':
+                value = get_act_layer('relu6')
+            elif v == 'hs':
+                value = get_act_layer('hard_swish')
+            elif v == 'sw':
+                value = get_act_layer('swish')
+            else:
+                continue
+            options[key] = value
+        else:
+            # all numeric options
+            splits = re.split(r'(\d.*)', op)
+            if len(splits) >= 2:
+                key, value = splits[:2]
+                options[key] = value
+
+    # if act_layer is None, the model default (passed to model init) will be used
+    act_layer = options['n'] if 'n' in options else None
+    exp_kernel_size = _parse_ksize(options['a']) if 'a' in options else 1
+    pw_kernel_size = _parse_ksize(options['p']) if 'p' in options else 1
+    fake_in_chs = int(options['fc']) if 'fc' in options else 0  # FIXME hack to deal with in_chs issue in TPU def
+
+    num_repeat = int(options['r'])
+    # each type of block has different valid arguments, fill accordingly
+    if block_type == 'ir':
+        block_args = dict(
+            block_type=block_type,
+            dw_kernel_size=_parse_ksize(options['k']),
+            exp_kernel_size=exp_kernel_size,
+            pw_kernel_size=pw_kernel_size,
+            out_chs=int(options['c']),
+            exp_ratio=float(options['e']),
+            se_ratio=float(options['se']) if 'se' in options else None,
+            stride=int(options['s']),
+            act_layer=act_layer,
+            noskip=noskip,
+        )
+        if 'cc' in options:
+            block_args['num_experts'] = int(options['cc'])
+    elif block_type == 'ds' or block_type == 'dsa':
+        block_args = dict(
+            block_type=block_type,
+            dw_kernel_size=_parse_ksize(options['k']),
+            pw_kernel_size=pw_kernel_size,
+            out_chs=int(options['c']),
+            se_ratio=float(options['se']) if 'se' in options else None,
+            stride=int(options['s']),
+            act_layer=act_layer,
+            pw_act=block_type == 'dsa',
+            noskip=block_type == 'dsa' or noskip,
+        )
+    elif block_type == 'er':
+        block_args = dict(
+            block_type=block_type,
+            exp_kernel_size=_parse_ksize(options['k']),
+            pw_kernel_size=pw_kernel_size,
+            out_chs=int(options['c']),
+            exp_ratio=float(options['e']),
+            fake_in_chs=fake_in_chs,
+            se_ratio=float(options['se']) if 'se' in options else None,
+            stride=int(options['s']),
+            act_layer=act_layer,
+            noskip=noskip,
+        )
+    elif block_type == 'cn':
+        block_args = dict(
+            block_type=block_type,
+            kernel_size=int(options['k']),
+            out_chs=int(options['c']),
+            stride=int(options['s']),
+            act_layer=act_layer,
+        )
+    else:
+        assert False, 'Unknown block type (%s)' % block_type
+
+    return block_args, num_repeat
+
+
+def _scale_stage_depth(stack_args, repeats, depth_multiplier=1.0, depth_trunc='ceil'):
+    """ Per-stage depth scaling
+    Scales the block repeats in each stage. This depth scaling impl maintains
+    compatibility with the EfficientNet scaling method, while allowing sensible
+    scaling for other models that may have multiple block arg definitions in each stage.
+    """
+
+    # We scale the total repeat count for each stage, there may be multiple
+    # block arg defs per stage so we need to sum.
+    num_repeat = sum(repeats)
+    if depth_trunc == 'round':
+        # Truncating to int by rounding allows stages with few repeats to remain
+        # proportionally smaller for longer. This is a good choice when stage definitions
+        # include single repeat stages that we'd prefer to keep that way as long as possible
+        num_repeat_scaled = max(1, round(num_repeat * depth_multiplier))
+    else:
+        # The default for EfficientNet truncates repeats to int via 'ceil'.
+        # Any multiplier > 1.0 will result in an increased depth for every stage.
+        num_repeat_scaled = int(math.ceil(num_repeat * depth_multiplier))
+
+    # Proportionally distribute repeat count scaling to each block definition in the stage.
+    # Allocation is done in reverse as it results in the first block being less likely to be scaled.
+    # The first block makes less sense to repeat in most of the arch definitions.
+    repeats_scaled = []
+    for r in repeats[::-1]:
+        rs = max(1, round((r / num_repeat * num_repeat_scaled)))
+        repeats_scaled.append(rs)
+        num_repeat -= r
+        num_repeat_scaled -= rs
+    repeats_scaled = repeats_scaled[::-1]
+
+    # Apply the calculated scaling to each block arg in the stage
+    sa_scaled = []
+    for ba, rep in zip(stack_args, repeats_scaled):
+        sa_scaled.extend([deepcopy(ba) for _ in range(rep)])
+    return sa_scaled
+
+
+def decode_arch_def(arch_def, depth_multiplier=1.0, depth_trunc='ceil', experts_multiplier=1, fix_first_last=False):
+    arch_args = []
+    for stack_idx, block_strings in enumerate(arch_def):
+        assert isinstance(block_strings, list)
+        stack_args = []
+        repeats = []
+        for block_str in block_strings:
+            assert isinstance(block_str, str)
+            ba, rep = _decode_block_str(block_str)
+            if ba.get('num_experts', 0) > 0 and experts_multiplier > 1:
+                ba['num_experts'] *= experts_multiplier
+            stack_args.append(ba)
+            repeats.append(rep)
+        if fix_first_last and (stack_idx == 0 or stack_idx == len(arch_def) - 1):
+            arch_args.append(_scale_stage_depth(stack_args, repeats, 1.0, depth_trunc))
+        else:
+            arch_args.append(_scale_stage_depth(stack_args, repeats, depth_multiplier, depth_trunc))
+    return arch_args
+
+
+def initialize_weight_goog(m, n='', fix_group_fanout=True):
+    # weight init as per Tensorflow Official impl
+    # https://github.com/tensorflow/tpu/blob/master/models/official/mnasnet/mnasnet_model.py
+    if isinstance(m, CondConv2d):
+        fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
+        if fix_group_fanout:
+            fan_out //= m.groups
+        init_weight_fn = get_condconv_initializer(
+            lambda w: w.data.normal_(0, math.sqrt(2.0 / fan_out)), m.num_experts, m.weight_shape)
+        init_weight_fn(m.weight)
+        if m.bias is not None:
+            m.bias.data.zero_()
+    elif isinstance(m, nn.Conv2d):
+        fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
+        if fix_group_fanout:
+            fan_out //= m.groups
+        m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
+        if m.bias is not None:
+            m.bias.data.zero_()
+    elif isinstance(m, nn.BatchNorm2d):
+        m.weight.data.fill_(1.0)
+        m.bias.data.zero_()
+    elif isinstance(m, nn.Linear):
+        fan_out = m.weight.size(0)  # fan-out
+        fan_in = 0
+        if 'routing_fn' in n:
+            fan_in = m.weight.size(1)
+        init_range = 1.0 / math.sqrt(fan_in + fan_out)
+        m.weight.data.uniform_(-init_range, init_range)
+        m.bias.data.zero_()
+
+
+def initialize_weight_default(m, n=''):
+    if isinstance(m, CondConv2d):
+        init_fn = get_condconv_initializer(partial(
+            nn.init.kaiming_normal_, mode='fan_out', nonlinearity='relu'), m.num_experts, m.weight_shape)
+        init_fn(m.weight)
+    elif isinstance(m, nn.Conv2d):
+        nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+    elif isinstance(m, nn.BatchNorm2d):
+        m.weight.data.fill_(1.0)
+        m.bias.data.zero_()
+    elif isinstance(m, nn.Linear):
+        nn.init.kaiming_uniform_(m.weight, mode='fan_in', nonlinearity='linear')

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/helpers.py

@@ -0,0 +1,71 @@
+""" Checkpoint loading / state_dict helpers
+Copyright 2020 Ross Wightman
+"""
+import torch
+import os
+from collections import OrderedDict
+try:
+    from torch.hub import load_state_dict_from_url
+except ImportError:
+    from torch.utils.model_zoo import load_url as load_state_dict_from_url
+
+
+def load_checkpoint(model, checkpoint_path):
+    if checkpoint_path and os.path.isfile(checkpoint_path):
+        print("=> Loading checkpoint '{}'".format(checkpoint_path))
+        checkpoint = torch.load(checkpoint_path)
+        if isinstance(checkpoint, dict) and 'state_dict' in checkpoint:
+            new_state_dict = OrderedDict()
+            for k, v in checkpoint['state_dict'].items():
+                if k.startswith('module'):
+                    name = k[7:]  # remove `module.`
+                else:
+                    name = k
+                new_state_dict[name] = v
+            model.load_state_dict(new_state_dict)
+        else:
+            model.load_state_dict(checkpoint)
+        print("=> Loaded checkpoint '{}'".format(checkpoint_path))
+    else:
+        print("=> Error: No checkpoint found at '{}'".format(checkpoint_path))
+        raise FileNotFoundError()
+
+
+def load_pretrained(model, url, filter_fn=None, strict=True):
+    if not url:
+        print("=> Warning: Pretrained model URL is empty, using random initialization.")
+        return
+
+    state_dict = load_state_dict_from_url(url, progress=False, map_location='cpu')
+
+    input_conv = 'conv_stem'
+    classifier = 'classifier'
+    in_chans = getattr(model, input_conv).weight.shape[1]
+    num_classes = getattr(model, classifier).weight.shape[0]
+
+    input_conv_weight = input_conv + '.weight'
+    pretrained_in_chans = state_dict[input_conv_weight].shape[1]
+    if in_chans != pretrained_in_chans:
+        if in_chans == 1:
+            print('=> Converting pretrained input conv {} from {} to 1 channel'.format(
+                input_conv_weight, pretrained_in_chans))
+            conv1_weight = state_dict[input_conv_weight]
+            state_dict[input_conv_weight] = conv1_weight.sum(dim=1, keepdim=True)
+        else:
+            print('=> Discarding pretrained input conv {} since input channel count != {}'.format(
+                input_conv_weight, pretrained_in_chans))
+            del state_dict[input_conv_weight]
+            strict = False
+
+    classifier_weight = classifier + '.weight'
+    pretrained_num_classes = state_dict[classifier_weight].shape[0]
+    if num_classes != pretrained_num_classes:
+        print('=> Discarding pretrained classifier since num_classes != {}'.format(pretrained_num_classes))
+        del state_dict[classifier_weight]
+        del state_dict[classifier + '.bias']
+        strict = False
+
+    if filter_fn is not None:
+        state_dict = filter_fn(state_dict)
+
+    model.load_state_dict(state_dict, strict=strict)

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/mobilenetv3.py

@@ -0,0 +1,364 @@
+""" MobileNet-V3
+
+A PyTorch impl of MobileNet-V3, compatible with TF weights from official impl.
+
+Paper: Searching for MobileNetV3 - https://arxiv.org/abs/1905.02244
+
+Hacked together by / Copyright 2020 Ross Wightman
+"""
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .activations import get_act_fn, get_act_layer, HardSwish
+from .config import layer_config_kwargs
+from .conv2d_layers import select_conv2d
+from .helpers import load_pretrained
+from .efficientnet_builder import *
+
+__all__ = ['mobilenetv3_rw', 'mobilenetv3_large_075', 'mobilenetv3_large_100', 'mobilenetv3_large_minimal_100',
+           'mobilenetv3_small_075', 'mobilenetv3_small_100', 'mobilenetv3_small_minimal_100',
+           'tf_mobilenetv3_large_075', 'tf_mobilenetv3_large_100', 'tf_mobilenetv3_large_minimal_100',
+           'tf_mobilenetv3_small_075', 'tf_mobilenetv3_small_100', 'tf_mobilenetv3_small_minimal_100']
+
+model_urls = {
+    'mobilenetv3_rw':
+        'https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mobilenetv3_100-35495452.pth',
+    'mobilenetv3_large_075': None,
+    'mobilenetv3_large_100':
+        'https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/mobilenetv3_large_100_ra-f55367f5.pth',
+    'mobilenetv3_large_minimal_100': None,
+    'mobilenetv3_small_075': None,
+    'mobilenetv3_small_100': None,
+    'mobilenetv3_small_minimal_100': None,
+    'tf_mobilenetv3_large_075':
+        'https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_mobilenetv3_large_075-150ee8b0.pth',
+    'tf_mobilenetv3_large_100':
+        'https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_mobilenetv3_large_100-427764d5.pth',
+    'tf_mobilenetv3_large_minimal_100':
+        'https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_mobilenetv3_large_minimal_100-8596ae28.pth',
+    'tf_mobilenetv3_small_075':
+        'https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_mobilenetv3_small_075-da427f52.pth',
+    'tf_mobilenetv3_small_100':
+        'https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_mobilenetv3_small_100-37f49e2b.pth',
+    'tf_mobilenetv3_small_minimal_100':
+        'https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/tf_mobilenetv3_small_minimal_100-922a7843.pth',
+}
+
+
+class MobileNetV3(nn.Module):
+    """ MobileNet-V3
+
+    A this model utilizes the MobileNet-v3 specific 'efficient head', where global pooling is done before the
+    head convolution without a final batch-norm layer before the classifier.
+
+    Paper: https://arxiv.org/abs/1905.02244
+    """
+
+    def __init__(self, block_args, num_classes=1000, in_chans=3, stem_size=16, num_features=1280, head_bias=True,
+                 channel_multiplier=1.0, pad_type='', act_layer=HardSwish, drop_rate=0., drop_connect_rate=0.,
+                 se_kwargs=None, norm_layer=nn.BatchNorm2d, norm_kwargs=None, weight_init='goog'):
+        super(MobileNetV3, self).__init__()
+        self.drop_rate = drop_rate
+
+        stem_size = round_channels(stem_size, channel_multiplier)
+        self.conv_stem = select_conv2d(in_chans, stem_size, 3, stride=2, padding=pad_type)
+        self.bn1 = nn.BatchNorm2d(stem_size, **norm_kwargs)
+        self.act1 = act_layer(inplace=True)
+        in_chs = stem_size
+
+        builder = EfficientNetBuilder(
+            channel_multiplier, pad_type=pad_type, act_layer=act_layer, se_kwargs=se_kwargs,
+            norm_layer=norm_layer, norm_kwargs=norm_kwargs, drop_connect_rate=drop_connect_rate)
+        self.blocks = nn.Sequential(*builder(in_chs, block_args))
+        in_chs = builder.in_chs
+
+        self.global_pool = nn.AdaptiveAvgPool2d(1)
+        self.conv_head = select_conv2d(in_chs, num_features, 1, padding=pad_type, bias=head_bias)
+        self.act2 = act_layer(inplace=True)
+        self.classifier = nn.Linear(num_features, num_classes)
+
+        for m in self.modules():
+            if weight_init == 'goog':
+                initialize_weight_goog(m)
+            else:
+                initialize_weight_default(m)
+
+    def as_sequential(self):
+        layers = [self.conv_stem, self.bn1, self.act1]
+        layers.extend(self.blocks)
+        layers.extend([
+            self.global_pool, self.conv_head, self.act2,
+            nn.Flatten(), nn.Dropout(self.drop_rate), self.classifier])
+        return nn.Sequential(*layers)
+
+    def features(self, x):
+        x = self.conv_stem(x)
+        x = self.bn1(x)
+        x = self.act1(x)
+        x = self.blocks(x)
+        x = self.global_pool(x)
+        x = self.conv_head(x)
+        x = self.act2(x)
+        return x
+
+    def forward(self, x):
+        x = self.features(x)
+        x = x.flatten(1)
+        if self.drop_rate > 0.:
+            x = F.dropout(x, p=self.drop_rate, training=self.training)
+        return self.classifier(x)
+
+
+def _create_model(model_kwargs, variant, pretrained=False):
+    as_sequential = model_kwargs.pop('as_sequential', False)
+    model = MobileNetV3(**model_kwargs)
+    if pretrained and model_urls[variant]:
+        load_pretrained(model, model_urls[variant])
+    if as_sequential:
+        model = model.as_sequential()
+    return model
+
+
+def _gen_mobilenet_v3_rw(variant, channel_multiplier=1.0, pretrained=False, **kwargs):
+    """Creates a MobileNet-V3 model (RW variant).
+
+    Paper: https://arxiv.org/abs/1905.02244
+
+    This was my first attempt at reproducing the MobileNet-V3 from paper alone. It came close to the
+    eventual Tensorflow reference impl but has a few differences:
+    1. This model has no bias on the head convolution
+    2. This model forces no residual (noskip) on the first DWS block, this is different than MnasNet
+    3. This model always uses ReLU for the SE activation layer, other models in the family inherit their act layer
+       from their parent block
+    4. This model does not enforce divisible by 8 limitation on the SE reduction channel count
+
+    Overall the changes are fairly minor and result in a very small parameter count difference and no
+    top-1/5
+
+    Args:
+      channel_multiplier: multiplier to number of channels per layer.
+    """
+    arch_def = [
+        # stage 0, 112x112 in
+        ['ds_r1_k3_s1_e1_c16_nre_noskip'],  # relu
+        # stage 1, 112x112 in
+        ['ir_r1_k3_s2_e4_c24_nre', 'ir_r1_k3_s1_e3_c24_nre'],  # relu
+        # stage 2, 56x56 in
+        ['ir_r3_k5_s2_e3_c40_se0.25_nre'],  # relu
+        # stage 3, 28x28 in
+        ['ir_r1_k3_s2_e6_c80', 'ir_r1_k3_s1_e2.5_c80', 'ir_r2_k3_s1_e2.3_c80'],  # hard-swish
+        # stage 4, 14x14in
+        ['ir_r2_k3_s1_e6_c112_se0.25'],  # hard-swish
+        # stage 5, 14x14in
+        ['ir_r3_k5_s2_e6_c160_se0.25'],  # hard-swish
+        # stage 6, 7x7 in
+        ['cn_r1_k1_s1_c960'],  # hard-swish
+    ]
+    with layer_config_kwargs(kwargs):
+        model_kwargs = dict(
+            block_args=decode_arch_def(arch_def),
+            head_bias=False,  # one of my mistakes
+            channel_multiplier=channel_multiplier,
+            act_layer=resolve_act_layer(kwargs, 'hard_swish'),
+            se_kwargs=dict(gate_fn=get_act_fn('hard_sigmoid'), reduce_mid=True),
+            norm_kwargs=resolve_bn_args(kwargs),
+            **kwargs,
+        )
+        model = _create_model(model_kwargs, variant, pretrained)
+    return model
+
+
+def _gen_mobilenet_v3(variant, channel_multiplier=1.0, pretrained=False, **kwargs):
+    """Creates a MobileNet-V3 large/small/minimal models.
+
+    Ref impl: https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet_v3.py
+    Paper: https://arxiv.org/abs/1905.02244
+
+    Args:
+      channel_multiplier: multiplier to number of channels per layer.
+    """
+    if 'small' in variant:
+        num_features = 1024
+        if 'minimal' in variant:
+            act_layer = 'relu'
+            arch_def = [
+                # stage 0, 112x112 in
+                ['ds_r1_k3_s2_e1_c16'],
+                # stage 1, 56x56 in
+                ['ir_r1_k3_s2_e4.5_c24', 'ir_r1_k3_s1_e3.67_c24'],
+                # stage 2, 28x28 in
+                ['ir_r1_k3_s2_e4_c40', 'ir_r2_k3_s1_e6_c40'],
+                # stage 3, 14x14 in
+                ['ir_r2_k3_s1_e3_c48'],
+                # stage 4, 14x14in
+                ['ir_r3_k3_s2_e6_c96'],
+                # stage 6, 7x7 in
+                ['cn_r1_k1_s1_c576'],
+            ]
+        else:
+            act_layer = 'hard_swish'
+            arch_def = [
+                # stage 0, 112x112 in
+                ['ds_r1_k3_s2_e1_c16_se0.25_nre'],  # relu
+                # stage 1, 56x56 in
+                ['ir_r1_k3_s2_e4.5_c24_nre', 'ir_r1_k3_s1_e3.67_c24_nre'],  # relu
+                # stage 2, 28x28 in
+                ['ir_r1_k5_s2_e4_c40_se0.25', 'ir_r2_k5_s1_e6_c40_se0.25'],  # hard-swish
+                # stage 3, 14x14 in
+                ['ir_r2_k5_s1_e3_c48_se0.25'],  # hard-swish
+                # stage 4, 14x14in
+                ['ir_r3_k5_s2_e6_c96_se0.25'],  # hard-swish
+                # stage 6, 7x7 in
+                ['cn_r1_k1_s1_c576'],  # hard-swish
+            ]
+    else:
+        num_features = 1280
+        if 'minimal' in variant:
+            act_layer = 'relu'
+            arch_def = [
+                # stage 0, 112x112 in
+                ['ds_r1_k3_s1_e1_c16'],
+                # stage 1, 112x112 in
+                ['ir_r1_k3_s2_e4_c24', 'ir_r1_k3_s1_e3_c24'],
+                # stage 2, 56x56 in
+                ['ir_r3_k3_s2_e3_c40'],
+                # stage 3, 28x28 in
+                ['ir_r1_k3_s2_e6_c80', 'ir_r1_k3_s1_e2.5_c80', 'ir_r2_k3_s1_e2.3_c80'],
+                # stage 4, 14x14in
+                ['ir_r2_k3_s1_e6_c112'],
+                # stage 5, 14x14in
+                ['ir_r3_k3_s2_e6_c160'],
+                # stage 6, 7x7 in
+                ['cn_r1_k1_s1_c960'],
+            ]
+        else:
+            act_layer = 'hard_swish'
+            arch_def = [
+                # stage 0, 112x112 in
+                ['ds_r1_k3_s1_e1_c16_nre'],  # relu
+                # stage 1, 112x112 in
+                ['ir_r1_k3_s2_e4_c24_nre', 'ir_r1_k3_s1_e3_c24_nre'],  # relu
+                # stage 2, 56x56 in
+                ['ir_r3_k5_s2_e3_c40_se0.25_nre'],  # relu
+                # stage 3, 28x28 in
+                ['ir_r1_k3_s2_e6_c80', 'ir_r1_k3_s1_e2.5_c80', 'ir_r2_k3_s1_e2.3_c80'],  # hard-swish
+                # stage 4, 14x14in
+                ['ir_r2_k3_s1_e6_c112_se0.25'],  # hard-swish
+                # stage 5, 14x14in
+                ['ir_r3_k5_s2_e6_c160_se0.25'],  # hard-swish
+                # stage 6, 7x7 in
+                ['cn_r1_k1_s1_c960'],  # hard-swish
+            ]
+    with layer_config_kwargs(kwargs):
+        model_kwargs = dict(
+            block_args=decode_arch_def(arch_def),
+            num_features=num_features,
+            stem_size=16,
+            channel_multiplier=channel_multiplier,
+            act_layer=resolve_act_layer(kwargs, act_layer),
+            se_kwargs=dict(
+                act_layer=get_act_layer('relu'), gate_fn=get_act_fn('hard_sigmoid'), reduce_mid=True, divisor=8),
+            norm_kwargs=resolve_bn_args(kwargs),
+            **kwargs,
+        )
+        model = _create_model(model_kwargs, variant, pretrained)
+    return model
+
+
+def mobilenetv3_rw(pretrained=False, **kwargs):
+    """ MobileNet-V3 RW
+    Attn: See note in gen function for this variant.
+    """
+    # NOTE for train set drop_rate=0.2
+    if pretrained:
+        # pretrained model trained with non-default BN epsilon
+        kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
+    model = _gen_mobilenet_v3_rw('mobilenetv3_rw', 1.0, pretrained=pretrained, **kwargs)
+    return model
+
+
+def mobilenetv3_large_075(pretrained=False, **kwargs):
+    """ MobileNet V3 Large 0.75"""
+    # NOTE for train set drop_rate=0.2
+    model = _gen_mobilenet_v3('mobilenetv3_large_075', 0.75, pretrained=pretrained, **kwargs)
+    return model
+
+
+def mobilenetv3_large_100(pretrained=False, **kwargs):
+    """ MobileNet V3 Large 1.0 """
+    # NOTE for train set drop_rate=0.2
+    model = _gen_mobilenet_v3('mobilenetv3_large_100', 1.0, pretrained=pretrained, **kwargs)
+    return model
+
+
+def mobilenetv3_large_minimal_100(pretrained=False, **kwargs):
+    """ MobileNet V3 Large (Minimalistic) 1.0 """
+    # NOTE for train set drop_rate=0.2
+    model = _gen_mobilenet_v3('mobilenetv3_large_minimal_100', 1.0, pretrained=pretrained, **kwargs)
+    return model
+
+
+def mobilenetv3_small_075(pretrained=False, **kwargs):
+    """ MobileNet V3 Small 0.75 """
+    model = _gen_mobilenet_v3('mobilenetv3_small_075', 0.75, pretrained=pretrained, **kwargs)
+    return model
+
+
+def mobilenetv3_small_100(pretrained=False, **kwargs):
+    """ MobileNet V3 Small 1.0 """
+    model = _gen_mobilenet_v3('mobilenetv3_small_100', 1.0, pretrained=pretrained, **kwargs)
+    return model
+
+
+def mobilenetv3_small_minimal_100(pretrained=False, **kwargs):
+    """ MobileNet V3 Small (Minimalistic) 1.0 """
+    model = _gen_mobilenet_v3('mobilenetv3_small_minimal_100', 1.0, pretrained=pretrained, **kwargs)
+    return model
+
+
+def tf_mobilenetv3_large_075(pretrained=False, **kwargs):
+    """ MobileNet V3 Large 0.75. Tensorflow compat variant. """
+    kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
+    kwargs['pad_type'] = 'same'
+    model = _gen_mobilenet_v3('tf_mobilenetv3_large_075', 0.75, pretrained=pretrained, **kwargs)
+    return model
+
+
+def tf_mobilenetv3_large_100(pretrained=False, **kwargs):
+    """ MobileNet V3 Large 1.0. Tensorflow compat variant. """
+    kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
+    kwargs['pad_type'] = 'same'
+    model = _gen_mobilenet_v3('tf_mobilenetv3_large_100', 1.0, pretrained=pretrained, **kwargs)
+    return model
+
+
+def tf_mobilenetv3_large_minimal_100(pretrained=False, **kwargs):
+    """ MobileNet V3 Large Minimalistic 1.0. Tensorflow compat variant. """
+    kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
+    kwargs['pad_type'] = 'same'
+    model = _gen_mobilenet_v3('tf_mobilenetv3_large_minimal_100', 1.0, pretrained=pretrained, **kwargs)
+    return model
+
+
+def tf_mobilenetv3_small_075(pretrained=False, **kwargs):
+    """ MobileNet V3 Small 0.75. Tensorflow compat variant. """
+    kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
+    kwargs['pad_type'] = 'same'
+    model = _gen_mobilenet_v3('tf_mobilenetv3_small_075', 0.75, pretrained=pretrained, **kwargs)
+    return model
+
+
+def tf_mobilenetv3_small_100(pretrained=False, **kwargs):
+    """ MobileNet V3 Small 1.0. Tensorflow compat variant."""
+    kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
+    kwargs['pad_type'] = 'same'
+    model = _gen_mobilenet_v3('tf_mobilenetv3_small_100', 1.0, pretrained=pretrained, **kwargs)
+    return model
+
+
+def tf_mobilenetv3_small_minimal_100(pretrained=False, **kwargs):
+    """ MobileNet V3 Small Minimalistic 1.0. Tensorflow compat variant. """
+    kwargs['bn_eps'] = BN_EPS_TF_DEFAULT
+    kwargs['pad_type'] = 'same'
+    model = _gen_mobilenet_v3('tf_mobilenetv3_small_minimal_100', 1.0, pretrained=pretrained, **kwargs)
+    return model

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/model_factory.py

@@ -0,0 +1,27 @@
+from .config import set_layer_config
+from .helpers import load_checkpoint
+
+from .gen_efficientnet import *
+from .mobilenetv3 import *
+
+
+def create_model(
+        model_name='mnasnet_100',
+        pretrained=None,
+        num_classes=1000,
+        in_chans=3,
+        checkpoint_path='',
+        **kwargs):
+
+    model_kwargs = dict(num_classes=num_classes, in_chans=in_chans, pretrained=pretrained, **kwargs)
+
+    if model_name in globals():
+        create_fn = globals()[model_name]
+        model = create_fn(**model_kwargs)
+    else:
+        raise RuntimeError('Unknown model (%s)' % model_name)
+
+    if checkpoint_path and not pretrained:
+        load_checkpoint(model, checkpoint_path)
+
+    return model

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/geffnet/version.py

@@ -0,0 +1 @@
+__version__ = '1.0.2'

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/hubconf.py

@@ -0,0 +1,84 @@
+dependencies = ['torch', 'math']
+
+from geffnet import efficientnet_b0
+from geffnet import efficientnet_b1
+from geffnet import efficientnet_b2
+from geffnet import efficientnet_b3
+
+from geffnet import efficientnet_es
+
+from geffnet import efficientnet_lite0
+
+from geffnet import mixnet_s
+from geffnet import mixnet_m
+from geffnet import mixnet_l
+from geffnet import mixnet_xl
+
+from geffnet import mobilenetv2_100
+from geffnet import mobilenetv2_110d
+from geffnet import mobilenetv2_120d
+from geffnet import mobilenetv2_140
+
+from geffnet import mobilenetv3_large_100
+from geffnet import mobilenetv3_rw
+from geffnet import mnasnet_a1
+from geffnet import mnasnet_b1
+from geffnet import fbnetc_100
+from geffnet import spnasnet_100
+
+from geffnet import tf_efficientnet_b0
+from geffnet import tf_efficientnet_b1
+from geffnet import tf_efficientnet_b2
+from geffnet import tf_efficientnet_b3
+from geffnet import tf_efficientnet_b4
+from geffnet import tf_efficientnet_b5
+from geffnet import tf_efficientnet_b6
+from geffnet import tf_efficientnet_b7
+from geffnet import tf_efficientnet_b8
+
+from geffnet import tf_efficientnet_b0_ap
+from geffnet import tf_efficientnet_b1_ap
+from geffnet import tf_efficientnet_b2_ap
+from geffnet import tf_efficientnet_b3_ap
+from geffnet import tf_efficientnet_b4_ap
+from geffnet import tf_efficientnet_b5_ap
+from geffnet import tf_efficientnet_b6_ap
+from geffnet import tf_efficientnet_b7_ap
+from geffnet import tf_efficientnet_b8_ap
+
+from geffnet import tf_efficientnet_b0_ns
+from geffnet import tf_efficientnet_b1_ns
+from geffnet import tf_efficientnet_b2_ns
+from geffnet import tf_efficientnet_b3_ns
+from geffnet import tf_efficientnet_b4_ns
+from geffnet import tf_efficientnet_b5_ns
+from geffnet import tf_efficientnet_b6_ns
+from geffnet import tf_efficientnet_b7_ns
+from geffnet import tf_efficientnet_l2_ns_475
+from geffnet import tf_efficientnet_l2_ns
+
+from geffnet import tf_efficientnet_es
+from geffnet import tf_efficientnet_em
+from geffnet import tf_efficientnet_el
+
+from geffnet import tf_efficientnet_cc_b0_4e
+from geffnet import tf_efficientnet_cc_b0_8e
+from geffnet import tf_efficientnet_cc_b1_8e
+
+from geffnet import tf_efficientnet_lite0
+from geffnet import tf_efficientnet_lite1
+from geffnet import tf_efficientnet_lite2
+from geffnet import tf_efficientnet_lite3
+from geffnet import tf_efficientnet_lite4
+
+from geffnet import tf_mixnet_s
+from geffnet import tf_mixnet_m
+from geffnet import tf_mixnet_l
+
+from geffnet import tf_mobilenetv3_large_075
+from geffnet import tf_mobilenetv3_large_100
+from geffnet import tf_mobilenetv3_large_minimal_100
+from geffnet import tf_mobilenetv3_small_075
+from geffnet import tf_mobilenetv3_small_100
+from geffnet import tf_mobilenetv3_small_minimal_100
+

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/onnx_export.py

@@ -0,0 +1,120 @@
+""" ONNX export script
+
+Export PyTorch models as ONNX graphs.
+
+This export script originally started as an adaptation of code snippets found at
+https://pytorch.org/tutorials/advanced/super_resolution_with_onnxruntime.html
+
+The default parameters work with PyTorch 1.6 and ONNX 1.7 and produce an optimal ONNX graph
+for hosting in the ONNX runtime (see onnx_validate.py). To export an ONNX model compatible
+with caffe2 (see caffe2_benchmark.py and caffe2_validate.py), the --keep-init and --aten-fallback
+flags are currently required.
+
+Older versions of PyTorch/ONNX (tested PyTorch 1.4, ONNX 1.5) do not need extra flags for
+caffe2 compatibility, but they produce a model that isn't as fast running on ONNX runtime.
+
+Most new release of PyTorch and ONNX cause some sort of breakage in the export / usage of ONNX models.
+Please do your research and search ONNX and PyTorch issue tracker before asking me. Thanks.
+
+Copyright 2020 Ross Wightman
+"""
+import argparse
+import torch
+import numpy as np
+
+import onnx
+import geffnet
+
+parser = argparse.ArgumentParser(description='PyTorch ImageNet Validation')
+parser.add_argument('output', metavar='ONNX_FILE',
+                    help='output model filename')
+parser.add_argument('--model', '-m', metavar='MODEL', default='mobilenetv3_large_100',
+                    help='model architecture (default: mobilenetv3_large_100)')
+parser.add_argument('--opset', type=int, default=10,
+                    help='ONNX opset to use (default: 10)')
+parser.add_argument('--keep-init', action='store_true', default=False,
+                    help='Keep initializers as input. Needed for Caffe2 compatible export in newer PyTorch/ONNX.')
+parser.add_argument('--aten-fallback', action='store_true', default=False,
+                    help='Fallback to ATEN ops. Helps fix AdaptiveAvgPool issue with Caffe2 in newer PyTorch/ONNX.')
+parser.add_argument('--dynamic-size', action='store_true', default=False,
+                    help='Export model width dynamic width/height. Not recommended for "tf" models with SAME padding.')
+parser.add_argument('-b', '--batch-size', default=1, type=int,
+                    metavar='N', help='mini-batch size (default: 1)')
+parser.add_argument('--img-size', default=None, type=int,
+                    metavar='N', help='Input image dimension, uses model default if empty')
+parser.add_argument('--mean', type=float, nargs='+', default=None, metavar='MEAN',
+                    help='Override mean pixel value of dataset')
+parser.add_argument('--std', type=float,  nargs='+', default=None, metavar='STD',
+                    help='Override std deviation of of dataset')
+parser.add_argument('--num-classes', type=int, default=1000,
+                    help='Number classes in dataset')
+parser.add_argument('--checkpoint', default='', type=str, metavar='PATH',
+                    help='path to checkpoint (default: none)')
+
+
+def main():
+    args = parser.parse_args()
+
+    args.pretrained = True
+    if args.checkpoint:
+        args.pretrained = False
+
+    print("==> Creating PyTorch {} model".format(args.model))
+    # NOTE exportable=True flag disables autofn/jit scripted activations and uses Conv2dSameExport layers
+    # for models using SAME padding
+    model = geffnet.create_model(
+        args.model,
+        num_classes=args.num_classes,
+        in_chans=3,
+        pretrained=args.pretrained,
+        checkpoint_path=args.checkpoint,
+        exportable=True)
+
+    model.eval()
+
+    example_input = torch.randn((args.batch_size, 3, args.img_size or 224, args.img_size or 224), requires_grad=True)
+
+    # Run model once before export trace, sets padding for models with Conv2dSameExport. This means
+    # that the padding for models with Conv2dSameExport (most models with tf_ prefix) is fixed for
+    # the input img_size specified in this script.
+    # Opset >= 11 should allow for dynamic padding, however I cannot get it to work due to
+    # issues in the tracing of the dynamic padding or errors attempting to export the model after jit
+    # scripting it (an approach that should work). Perhaps in a future PyTorch or ONNX versions...
+    model(example_input)
+
+    print("==> Exporting model to ONNX format at '{}'".format(args.output))
+    input_names = ["input0"]
+    output_names = ["output0"]
+    dynamic_axes = {'input0': {0: 'batch'}, 'output0': {0: 'batch'}}
+    if args.dynamic_size:
+        dynamic_axes['input0'][2] = 'height'
+        dynamic_axes['input0'][3] = 'width'
+    if args.aten_fallback:
+        export_type = torch.onnx.OperatorExportTypes.ONNX_ATEN_FALLBACK
+    else:
+        export_type = torch.onnx.OperatorExportTypes.ONNX
+
+    torch_out = torch.onnx._export(
+        model, example_input, args.output, export_params=True, verbose=True, input_names=input_names,
+        output_names=output_names, keep_initializers_as_inputs=args.keep_init, dynamic_axes=dynamic_axes,
+        opset_version=args.opset, operator_export_type=export_type)
+
+    print("==> Loading and checking exported model from '{}'".format(args.output))
+    onnx_model = onnx.load(args.output)
+    onnx.checker.check_model(onnx_model)  # assuming throw on error
+    print("==> Passed")
+
+    if args.keep_init and args.aten_fallback:
+        import caffe2.python.onnx.backend as onnx_caffe2
+        # Caffe2 loading only works properly in newer PyTorch/ONNX combos when
+        # keep_initializers_as_inputs and aten_fallback are set to True.
+        print("==> Loading model into Caffe2 backend and comparing forward pass.".format(args.output))
+        caffe2_backend = onnx_caffe2.prepare(onnx_model)
+        B = {onnx_model.graph.input[0].name: x.data.numpy()}
+        c2_out = caffe2_backend.run(B)[0]
+        np.testing.assert_almost_equal(torch_out.data.numpy(), c2_out, decimal=5)
+        print("==> Passed")
+
+
+if __name__ == '__main__':
+    main()

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/onnx_optimize.py

@@ -0,0 +1,84 @@
+""" ONNX optimization script
+
+Run ONNX models through the optimizer to prune unneeded nodes, fuse batchnorm layers into conv, etc.
+
+NOTE: This isn't working consistently in recent PyTorch/ONNX combos (ie PyTorch 1.6 and ONNX 1.7),
+it seems time to switch to using the onnxruntime online optimizer (can also be saved for offline).
+
+Copyright 2020 Ross Wightman
+"""
+import argparse
+import warnings
+
+import onnx
+from onnx import optimizer
+
+
+parser = argparse.ArgumentParser(description="Optimize ONNX model")
+
+parser.add_argument("model", help="The ONNX model")
+parser.add_argument("--output", required=True, help="The optimized model output filename")
+
+
+def traverse_graph(graph, prefix=''):
+    content = []
+    indent = prefix + '  '
+    graphs = []
+    num_nodes = 0
+    for node in graph.node:
+        pn, gs = onnx.helper.printable_node(node, indent, subgraphs=True)
+        assert isinstance(gs, list)
+        content.append(pn)
+        graphs.extend(gs)
+        num_nodes += 1
+    for g in graphs:
+        g_count, g_str = traverse_graph(g)
+        content.append('\n' + g_str)
+        num_nodes += g_count
+    return num_nodes, '\n'.join(content)
+
+
+def main():
+    args = parser.parse_args()
+    onnx_model = onnx.load(args.model)
+    num_original_nodes, original_graph_str = traverse_graph(onnx_model.graph)
+
+    # Optimizer passes to perform
+    passes = [
+        #'eliminate_deadend',
+        'eliminate_identity',
+        'eliminate_nop_dropout',
+        'eliminate_nop_pad',
+        'eliminate_nop_transpose',
+        'eliminate_unused_initializer',
+        'extract_constant_to_initializer',
+        'fuse_add_bias_into_conv',
+        'fuse_bn_into_conv',
+        'fuse_consecutive_concats',
+        'fuse_consecutive_reduce_unsqueeze',
+        'fuse_consecutive_squeezes',
+        'fuse_consecutive_transposes',
+        #'fuse_matmul_add_bias_into_gemm',
+        'fuse_pad_into_conv',
+        #'fuse_transpose_into_gemm',
+        #'lift_lexical_references',
+    ]
+
+    # Apply the optimization on the original serialized model
+    # WARNING I've had issues with optimizer in recent versions of PyTorch / ONNX causing
+    # 'duplicate definition of name' errors, see: https://github.com/onnx/onnx/issues/2401
+    # It may be better to rely on onnxruntime optimizations, see onnx_validate.py script.
+    warnings.warn("I've had issues with optimizer in recent versions of PyTorch / ONNX."
+                  "Try onnxruntime optimization if this doesn't work.")
+    optimized_model = optimizer.optimize(onnx_model, passes)
+
+    num_optimized_nodes, optimzied_graph_str = traverse_graph(optimized_model.graph)
+    print('==> The model after optimization:\n{}\n'.format(optimzied_graph_str))
+    print('==> The optimized model has {} nodes, the original had {}.'.format(num_optimized_nodes, num_original_nodes))
+
+    # Save the ONNX model
+    onnx.save(optimized_model, args.output)
+
+
+if __name__ == "__main__":
+    main()

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/onnx_to_caffe.py

@@ -0,0 +1,27 @@
+import argparse
+
+import onnx
+from caffe2.python.onnx.backend import Caffe2Backend
+
+
+parser = argparse.ArgumentParser(description="Convert ONNX to Caffe2")
+
+parser.add_argument("model", help="The ONNX model")
+parser.add_argument("--c2-prefix", required=True,
+    help="The output file prefix for the caffe2 model init and predict file. ")
+
+
+def main():
+    args = parser.parse_args()
+    onnx_model = onnx.load(args.model)
+    caffe2_init, caffe2_predict = Caffe2Backend.onnx_graph_to_caffe2_net(onnx_model)
+    caffe2_init_str = caffe2_init.SerializeToString()
+    with open(args.c2_prefix + '.init.pb', "wb") as f:
+        f.write(caffe2_init_str)
+    caffe2_predict_str = caffe2_predict.SerializeToString()
+    with open(args.c2_prefix + '.predict.pb', "wb") as f:
+        f.write(caffe2_predict_str)
+
+
+if __name__ == "__main__":
+    main()

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/onnx_validate.py

@@ -0,0 +1,112 @@
+""" ONNX-runtime validation script
+
+This script was created to verify accuracy and performance of exported ONNX
+models running with the onnxruntime. It utilizes the PyTorch dataloader/processing
+pipeline for a fair comparison against the originals.
+
+Copyright 2020 Ross Wightman
+"""
+import argparse
+import numpy as np
+import onnxruntime
+from data import create_loader, resolve_data_config, Dataset
+from utils import AverageMeter
+import time
+
+parser = argparse.ArgumentParser(description='Caffe2 ImageNet Validation')
+parser.add_argument('data', metavar='DIR',
+                    help='path to dataset')
+parser.add_argument('--onnx-input', default='', type=str, metavar='PATH',
+                    help='path to onnx model/weights file')
+parser.add_argument('--onnx-output-opt', default='', type=str, metavar='PATH',
+                    help='path to output optimized onnx graph')
+parser.add_argument('--profile', action='store_true', default=False,
+                    help='Enable profiler output.')
+parser.add_argument('-j', '--workers', default=2, type=int, metavar='N',
+                    help='number of data loading workers (default: 2)')
+parser.add_argument('-b', '--batch-size', default=256, type=int,
+                    metavar='N', help='mini-batch size (default: 256)')
+parser.add_argument('--img-size', default=None, type=int,
+                    metavar='N', help='Input image dimension, uses model default if empty')
+parser.add_argument('--mean', type=float, nargs='+', default=None, metavar='MEAN',
+                    help='Override mean pixel value of dataset')
+parser.add_argument('--std', type=float,  nargs='+', default=None, metavar='STD',
+                    help='Override std deviation of of dataset')
+parser.add_argument('--crop-pct', type=float, default=None, metavar='PCT',
+                    help='Override default crop pct of 0.875')
+parser.add_argument('--interpolation', default='', type=str, metavar='NAME',
+                    help='Image resize interpolation type (overrides model)')
+parser.add_argument('--tf-preprocessing', dest='tf_preprocessing', action='store_true',
+                    help='use tensorflow mnasnet preporcessing')
+parser.add_argument('--print-freq', '-p', default=10, type=int,
+                    metavar='N', help='print frequency (default: 10)')
+
+
+def main():
+    args = parser.parse_args()
+    args.gpu_id = 0
+
+    # Set graph optimization level
+    sess_options = onnxruntime.SessionOptions()
+    sess_options.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL
+    if args.profile:
+        sess_options.enable_profiling = True
+    if args.onnx_output_opt:
+        sess_options.optimized_model_filepath = args.onnx_output_opt
+
+    session = onnxruntime.InferenceSession(args.onnx_input, sess_options)
+
+    data_config = resolve_data_config(None, args)
+    loader = create_loader(
+        Dataset(args.data, load_bytes=args.tf_preprocessing),
+        input_size=data_config['input_size'],
+        batch_size=args.batch_size,
+        use_prefetcher=False,
+        interpolation=data_config['interpolation'],
+        mean=data_config['mean'],
+        std=data_config['std'],
+        num_workers=args.workers,
+        crop_pct=data_config['crop_pct'],
+        tensorflow_preprocessing=args.tf_preprocessing)
+
+    input_name = session.get_inputs()[0].name
+
+    batch_time = AverageMeter()
+    top1 = AverageMeter()
+    top5 = AverageMeter()
+    end = time.time()
+    for i, (input, target) in enumerate(loader):
+        # run the net and return prediction
+        output = session.run([], {input_name: input.data.numpy()})
+        output = output[0]
+
+        # measure accuracy and record loss
+        prec1, prec5 = accuracy_np(output, target.numpy())
+        top1.update(prec1.item(), input.size(0))
+        top5.update(prec5.item(), input.size(0))
+
+        # measure elapsed time
+        batch_time.update(time.time() - end)
+        end = time.time()
+
+        if i % args.print_freq == 0:
+            print('Test: [{0}/{1}]\t'
+                  'Time {batch_time.val:.3f} ({batch_time.avg:.3f}, {rate_avg:.3f}/s, {ms_avg:.3f} ms/sample) \t'
+                  'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
+                  'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
+                i, len(loader), batch_time=batch_time, rate_avg=input.size(0) / batch_time.avg,
+                ms_avg=100 * batch_time.avg / input.size(0), top1=top1, top5=top5))
+
+    print(' * Prec@1 {top1.avg:.3f} ({top1a:.3f}) Prec@5 {top5.avg:.3f} ({top5a:.3f})'.format(
+        top1=top1, top1a=100-top1.avg, top5=top5, top5a=100.-top5.avg))
+
+
+def accuracy_np(output, target):
+    max_indices = np.argsort(output, axis=1)[:, ::-1]
+    top5 = 100 * np.equal(max_indices[:, :5], target[:, np.newaxis]).sum(axis=1).mean()
+    top1 = 100 * np.equal(max_indices[:, 0], target).mean()
+    return top1, top5
+
+
+if __name__ == '__main__':
+    main()

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/requirements.txt

@@ -0,0 +1,2 @@
+torch>=1.2.0
+torchvision>=0.4.0

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/setup.py

@@ -0,0 +1,47 @@
+""" Setup
+"""
+from setuptools import setup, find_packages
+from codecs import open
+from os import path
+
+here = path.abspath(path.dirname(__file__))
+
+# Get the long description from the README file
+with open(path.join(here, 'README.md'), encoding='utf-8') as f:
+    long_description = f.read()
+
+exec(open('geffnet/version.py').read())
+setup(
+    name='geffnet',
+    version=__version__,
+    description='(Generic) EfficientNets for PyTorch',
+    long_description=long_description,
+    long_description_content_type='text/markdown',
+    url='https://github.com/rwightman/gen-efficientnet-pytorch',
+    author='Ross Wightman',
+    author_email='hello@rwightman.com',
+    classifiers=[
+        # How mature is this project? Common values are
+        #   3 - Alpha
+        #   4 - Beta
+        #   5 - Production/Stable
+        'Development Status :: 3 - Alpha',
+        'Intended Audience :: Education',
+        'Intended Audience :: Science/Research',
+        'License :: OSI Approved :: Apache Software License',
+        'Programming Language :: Python :: 3.6',
+        'Programming Language :: Python :: 3.7',
+        'Programming Language :: Python :: 3.8',
+        'Topic :: Scientific/Engineering',
+        'Topic :: Scientific/Engineering :: Artificial Intelligence',
+        'Topic :: Software Development',
+        'Topic :: Software Development :: Libraries',
+        'Topic :: Software Development :: Libraries :: Python Modules',
+    ],
+
+    # Note that this is a string of words separated by whitespace, not a list.
+    keywords='pytorch pretrained models efficientnet mixnet mobilenetv3 mnasnet',
+    packages=find_packages(exclude=['data']),
+    install_requires=['torch >= 1.4', 'torchvision'],
+    python_requires='>=3.6',
+)

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/utils.py

@@ -0,0 +1,52 @@
+import os
+
+
+class AverageMeter:
+    """Computes and stores the average and current value"""
+    def __init__(self):
+        self.reset()
+
+    def reset(self):
+        self.val = 0
+        self.avg = 0
+        self.sum = 0
+        self.count = 0
+
+    def update(self, val, n=1):
+        self.val = val
+        self.sum += val * n
+        self.count += n
+        self.avg = self.sum / self.count
+
+
+def accuracy(output, target, topk=(1,)):
+    """Computes the precision@k for the specified values of k"""
+    maxk = max(topk)
+    batch_size = target.size(0)
+
+    _, pred = output.topk(maxk, 1, True, True)
+    pred = pred.t()
+    correct = pred.eq(target.view(1, -1).expand_as(pred))
+
+    res = []
+    for k in topk:
+        correct_k = correct[:k].reshape(-1).float().sum(0)
+        res.append(correct_k.mul_(100.0 / batch_size))
+    return res
+
+
+def get_outdir(path, *paths, inc=False):
+    outdir = os.path.join(path, *paths)
+    if not os.path.exists(outdir):
+        os.makedirs(outdir)
+    elif inc:
+        count = 1
+        outdir_inc = outdir + '-' + str(count)
+        while os.path.exists(outdir_inc):
+            count = count + 1
+            outdir_inc = outdir + '-' + str(count)
+            assert count < 100
+        outdir = outdir_inc
+        os.makedirs(outdir)
+    return outdir
+

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/efficientnet_repo/validate.py

@@ -0,0 +1,166 @@
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import argparse
+import time
+import torch
+import torch.nn as nn
+import torch.nn.parallel
+from contextlib import suppress
+
+import geffnet
+from data import Dataset, create_loader, resolve_data_config
+from utils import accuracy, AverageMeter
+
+has_native_amp = False
+try:
+    if getattr(torch.cuda.amp, 'autocast') is not None:
+        has_native_amp = True
+except AttributeError:
+    pass
+
+torch.backends.cudnn.benchmark = True
+
+parser = argparse.ArgumentParser(description='PyTorch ImageNet Validation')
+parser.add_argument('data', metavar='DIR',
+                    help='path to dataset')
+parser.add_argument('--model', '-m', metavar='MODEL', default='spnasnet1_00',
+                    help='model architecture (default: dpn92)')
+parser.add_argument('-j', '--workers', default=4, type=int, metavar='N',
+                    help='number of data loading workers (default: 2)')
+parser.add_argument('-b', '--batch-size', default=256, type=int,
+                    metavar='N', help='mini-batch size (default: 256)')
+parser.add_argument('--img-size', default=None, type=int,
+                    metavar='N', help='Input image dimension, uses model default if empty')
+parser.add_argument('--mean', type=float, nargs='+', default=None, metavar='MEAN',
+                    help='Override mean pixel value of dataset')
+parser.add_argument('--std', type=float,  nargs='+', default=None, metavar='STD',
+                    help='Override std deviation of of dataset')
+parser.add_argument('--crop-pct', type=float, default=None, metavar='PCT',
+                    help='Override default crop pct of 0.875')
+parser.add_argument('--interpolation', default='', type=str, metavar='NAME',
+                    help='Image resize interpolation type (overrides model)')
+parser.add_argument('--num-classes', type=int, default=1000,
+                    help='Number classes in dataset')
+parser.add_argument('--print-freq', '-p', default=10, type=int,
+                    metavar='N', help='print frequency (default: 10)')
+parser.add_argument('--checkpoint', default='', type=str, metavar='PATH',
+                    help='path to latest checkpoint (default: none)')
+parser.add_argument('--pretrained', dest='pretrained', action='store_true',
+                    help='use pre-trained model')
+parser.add_argument('--torchscript', dest='torchscript', action='store_true',
+                    help='convert model torchscript for inference')
+parser.add_argument('--num-gpu', type=int, default=1,
+                    help='Number of GPUS to use')
+parser.add_argument('--tf-preprocessing', dest='tf_preprocessing', action='store_true',
+                    help='use tensorflow mnasnet preporcessing')
+parser.add_argument('--no-cuda', dest='no_cuda', action='store_true',
+                    help='')
+parser.add_argument('--channels-last', action='store_true', default=False,
+                    help='Use channels_last memory layout')
+parser.add_argument('--amp', action='store_true', default=False,
+                    help='Use native Torch AMP mixed precision.')
+
+
+def main():
+    args = parser.parse_args()
+
+    if not args.checkpoint and not args.pretrained:
+        args.pretrained = True
+
+    amp_autocast = suppress  # do nothing
+    if args.amp:
+        if not has_native_amp:
+            print("Native Torch AMP is not available (requires torch >= 1.6), using FP32.")
+        else:
+            amp_autocast = torch.cuda.amp.autocast
+
+    # create model
+    model = geffnet.create_model(
+        args.model,
+        num_classes=args.num_classes,
+        in_chans=3,
+        pretrained=args.pretrained,
+        checkpoint_path=args.checkpoint,
+        scriptable=args.torchscript)
+
+    if args.channels_last:
+        model = model.to(memory_format=torch.channels_last)
+
+    if args.torchscript:
+        torch.jit.optimized_execution(True)
+        model = torch.jit.script(model)
+
+    print('Model %s created, param count: %d' %
+          (args.model, sum([m.numel() for m in model.parameters()])))
+
+    data_config = resolve_data_config(model, args)
+
+    criterion = nn.CrossEntropyLoss()
+
+    if not args.no_cuda:
+        if args.num_gpu > 1:
+            model = torch.nn.DataParallel(model, device_ids=list(range(args.num_gpu))).cuda()
+        else:
+            model = model.cuda()
+        criterion = criterion.cuda()
+
+    loader = create_loader(
+        Dataset(args.data, load_bytes=args.tf_preprocessing),
+        input_size=data_config['input_size'],
+        batch_size=args.batch_size,
+        use_prefetcher=not args.no_cuda,
+        interpolation=data_config['interpolation'],
+        mean=data_config['mean'],
+        std=data_config['std'],
+        num_workers=args.workers,
+        crop_pct=data_config['crop_pct'],
+        tensorflow_preprocessing=args.tf_preprocessing)
+
+    batch_time = AverageMeter()
+    losses = AverageMeter()
+    top1 = AverageMeter()
+    top5 = AverageMeter()
+
+    model.eval()
+    end = time.time()
+    with torch.no_grad():
+        for i, (input, target) in enumerate(loader):
+            if not args.no_cuda:
+                target = target.cuda()
+                input = input.cuda()
+            if args.channels_last:
+                input = input.contiguous(memory_format=torch.channels_last)
+
+            # compute output
+            with amp_autocast():
+                output = model(input)
+                loss = criterion(output, target)
+
+            # measure accuracy and record loss
+            prec1, prec5 = accuracy(output.data, target, topk=(1, 5))
+            losses.update(loss.item(), input.size(0))
+            top1.update(prec1.item(), input.size(0))
+            top5.update(prec5.item(), input.size(0))
+
+            # measure elapsed time
+            batch_time.update(time.time() - end)
+            end = time.time()
+
+            if i % args.print_freq == 0:
+                print('Test: [{0}/{1}]\t'
+                      'Time {batch_time.val:.3f} ({batch_time.avg:.3f}, {rate_avg:.3f}/s) \t'
+                      'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
+                      'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
+                      'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'.format(
+                    i, len(loader), batch_time=batch_time,
+                    rate_avg=input.size(0) / batch_time.avg,
+                    loss=losses, top1=top1, top5=top5))
+
+    print(' * Prec@1 {top1.avg:.3f} ({top1a:.3f}) Prec@5 {top5.avg:.3f} ({top5a:.3f})'.format(
+        top1=top1, top1a=100-top1.avg, top5=top5, top5a=100.-top5.avg))
+
+
+if __name__ == '__main__':
+    main()

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/encoder.py

@@ -0,0 +1,34 @@
+import os
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class Encoder(nn.Module):
+    def __init__(self):
+        super(Encoder, self).__init__()
+
+        basemodel_name = 'tf_efficientnet_b5_ap'
+        print('Loading base model ()...'.format(basemodel_name), end='')
+        repo_path = os.path.join(os.path.dirname(__file__), 'efficientnet_repo')
+        basemodel = torch.hub.load(repo_path, basemodel_name, pretrained=False, source='local')
+        print('Done.')
+
+        # Remove last layer
+        print('Removing last two layers (global_pool & classifier).')
+        basemodel.global_pool = nn.Identity()
+        basemodel.classifier = nn.Identity()
+
+        self.original_model = basemodel
+
+    def forward(self, x):
+        features = [x]
+        for k, v in self.original_model._modules.items():
+            if (k == 'blocks'):
+                for ki, vi in v._modules.items():
+                    features.append(vi(features[-1]))
+            else:
+                features.append(v(features[-1]))
+        return features
+
+

negi/repos/controlnet_aux/src/controlnet_aux/normalbae/nets/submodules/submodules.py

@@ -0,0 +1,140 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+########################################################################################################################
+
+
+# Upsample + BatchNorm
+class UpSampleBN(nn.Module):
+    def __init__(self, skip_input, output_features):
+        super(UpSampleBN, self).__init__()
+
+        self._net = nn.Sequential(nn.Conv2d(skip_input, output_features, kernel_size=3, stride=1, padding=1),
+                                  nn.BatchNorm2d(output_features),
+                                  nn.LeakyReLU(),
+                                  nn.Conv2d(output_features, output_features, kernel_size=3, stride=1, padding=1),
+                                  nn.BatchNorm2d(output_features),
+                                  nn.LeakyReLU())
+
+    def forward(self, x, concat_with):
+        up_x = F.interpolate(x, size=[concat_with.size(2), concat_with.size(3)], mode='bilinear', align_corners=True)
+        f = torch.cat([up_x, concat_with], dim=1)
+        return self._net(f)
+
+
+# Upsample + GroupNorm + Weight Standardization
+class UpSampleGN(nn.Module):
+    def __init__(self, skip_input, output_features):
+        super(UpSampleGN, self).__init__()
+
+        self._net = nn.Sequential(Conv2d(skip_input, output_features, kernel_size=3, stride=1, padding=1),
+                                  nn.GroupNorm(8, output_features),
+                                  nn.LeakyReLU(),
+                                  Conv2d(output_features, output_features, kernel_size=3, stride=1, padding=1),
+                                  nn.GroupNorm(8, output_features),
+                                  nn.LeakyReLU())
+
+    def forward(self, x, concat_with):
+        up_x = F.interpolate(x, size=[concat_with.size(2), concat_with.size(3)], mode='bilinear', align_corners=True)
+        f = torch.cat([up_x, concat_with], dim=1)
+        return self._net(f)
+
+
+# Conv2d with weight standardization
+class Conv2d(nn.Conv2d):
+    def __init__(self, in_channels, out_channels, kernel_size, stride=1,
+                 padding=0, dilation=1, groups=1, bias=True):
+        super(Conv2d, self).__init__(in_channels, out_channels, kernel_size, stride,
+                 padding, dilation, groups, bias)
+
+    def forward(self, x):
+        weight = self.weight
+        weight_mean = weight.mean(dim=1, keepdim=True).mean(dim=2,
+                                  keepdim=True).mean(dim=3, keepdim=True)
+        weight = weight - weight_mean
+        std = weight.view(weight.size(0), -1).std(dim=1).view(-1, 1, 1, 1) + 1e-5
+        weight = weight / std.expand_as(weight)
+        return F.conv2d(x, weight, self.bias, self.stride,
+                        self.padding, self.dilation, self.groups)
+
+
+# normalize
+def norm_normalize(norm_out):
+    min_kappa = 0.01
+    norm_x, norm_y, norm_z, kappa = torch.split(norm_out, 1, dim=1)
+    norm = torch.sqrt(norm_x ** 2.0 + norm_y ** 2.0 + norm_z ** 2.0) + 1e-10
+    kappa = F.elu(kappa) + 1.0 + min_kappa
+    final_out = torch.cat([norm_x / norm, norm_y / norm, norm_z / norm, kappa], dim=1)
+    return final_out
+
+
+# uncertainty-guided sampling (only used during training)
+@torch.no_grad()
+def sample_points(init_normal, gt_norm_mask, sampling_ratio, beta):
+    device = init_normal.device
+    B, _, H, W = init_normal.shape
+    N = int(sampling_ratio * H * W)
+    beta = beta
+
+    # uncertainty map
+    uncertainty_map = -1 * init_normal[:, 3, :, :]  # B, H, W
+
+    # gt_invalid_mask (B, H, W)
+    if gt_norm_mask is not None:
+        gt_invalid_mask = F.interpolate(gt_norm_mask.float(), size=[H, W], mode='nearest')
+        gt_invalid_mask = gt_invalid_mask[:, 0, :, :] < 0.5
+        uncertainty_map[gt_invalid_mask] = -1e4
+
+    # (B, H*W)
+    _, idx = uncertainty_map.view(B, -1).sort(1, descending=True)
+
+    # importance sampling
+    if int(beta * N) > 0:
+        importance = idx[:, :int(beta * N)]    # B, beta*N
+
+        # remaining
+        remaining = idx[:, int(beta * N):]     # B, H*W - beta*N
+
+        # coverage
+        num_coverage = N - int(beta * N)
+
+        if num_coverage <= 0:
+            samples = importance
+        else:
+            coverage_list = []
+            for i in range(B):
+                idx_c = torch.randperm(remaining.size()[1])    # shuffles "H*W - beta*N"
+                coverage_list.append(remaining[i, :][idx_c[:num_coverage]].view(1, -1))     # 1, N-beta*N
+            coverage = torch.cat(coverage_list, dim=0)                                      # B, N-beta*N
+            samples = torch.cat((importance, coverage), dim=1)                              # B, N
+
+    else:
+        # remaining
+        remaining = idx[:, :]  # B, H*W
+
+        # coverage
+        num_coverage = N
+
+        coverage_list = []
+        for i in range(B):
+            idx_c = torch.randperm(remaining.size()[1])  # shuffles "H*W - beta*N"
+            coverage_list.append(remaining[i, :][idx_c[:num_coverage]].view(1, -1))  # 1, N-beta*N
+        coverage = torch.cat(coverage_list, dim=0)  # B, N-beta*N
+        samples = coverage
+
+    # point coordinates
+    rows_int = samples // W         # 0 for first row, H-1 for last row
+    rows_float = rows_int / float(H-1)         # 0 to 1.0
+    rows_float = (rows_float * 2.0) - 1.0       # -1.0 to 1.0
+
+    cols_int = samples % W          # 0 for first column, W-1 for last column
+    cols_float = cols_int / float(W-1)         # 0 to 1.0
+    cols_float = (cols_float * 2.0) - 1.0       # -1.0 to 1.0
+
+    point_coords = torch.zeros(B, 1, N, 2)
+    point_coords[:, 0, :, 0] = cols_float             # x coord
+    point_coords[:, 0, :, 1] = rows_float             # y coord
+    point_coords = point_coords.to(device)
+    return point_coords, rows_int, cols_int

negi/repos/controlnet_aux/src/controlnet_aux/open_pose/LICENSE

@@ -0,0 +1,108 @@
+OPENPOSE: MULTIPERSON KEYPOINT DETECTION
+SOFTWARE LICENSE AGREEMENT
+ACADEMIC OR NON-PROFIT ORGANIZATION NONCOMMERCIAL RESEARCH USE ONLY
+
+BY USING OR DOWNLOADING THE SOFTWARE, YOU ARE AGREEING TO THE TERMS OF THIS LICENSE AGREEMENT.  IF YOU DO NOT AGREE WITH THESE TERMS, YOU MAY NOT USE OR DOWNLOAD THE SOFTWARE.
+
+This is a license agreement ("Agreement") between your academic institution or non-profit organization or self (called "Licensee" or "You" in this Agreement) and Carnegie Mellon University (called "Licensor" in this Agreement).  All rights not specifically granted to you in this Agreement are reserved for Licensor. 
+
+RESERVATION OF OWNERSHIP AND GRANT OF LICENSE: 
+Licensor retains exclusive ownership of any copy of the Software (as defined below) licensed under this Agreement and hereby grants to Licensee a personal, non-exclusive, 
+non-transferable license to use the Software for noncommercial research purposes, without the right to sublicense, pursuant to the terms and conditions of this Agreement.  As used in this Agreement, the term "Software" means (i) the actual copy of all or any portion of code for program routines made accessible to Licensee by Licensor pursuant to this Agreement, inclusive of backups, updates, and/or merged copies permitted hereunder or subsequently supplied by Licensor,  including all or any file structures, programming instructions, user interfaces and screen formats and sequences as well as any and all documentation and instructions related to it, and (ii) all or any derivatives and/or modifications created or made by You to any of the items specified in (i).
+
+CONFIDENTIALITY: Licensee acknowledges that the Software is proprietary to Licensor, and as such, Licensee agrees to receive all such materials in confidence and use the Software only in accordance with the terms of this Agreement.  Licensee agrees to use reasonable effort to protect the Software from unauthorized use, reproduction, distribution, or publication.
+
+COPYRIGHT: The Software is owned by Licensor and is protected by United 
+States copyright laws and applicable international treaties and/or conventions.
+
+PERMITTED USES:  The Software may be used for your own noncommercial internal research purposes. You understand and agree that Licensor is not obligated to implement any suggestions and/or feedback you might provide regarding the Software, but to the extent Licensor does so, you are not entitled to any compensation related thereto.
+
+DERIVATIVES: You may create derivatives of or make modifications to the Software, however, You agree that all and any such derivatives and modifications will be owned by Licensor and become a part of the Software licensed to You under this Agreement.  You may only use such derivatives and modifications for your own noncommercial internal research purposes, and you may not otherwise use, distribute or copy such derivatives and modifications in violation of this Agreement.
+
+BACKUPS:  If Licensee is an organization, it may make that number of copies of the Software necessary for internal noncommercial use at a single site within its organization provided that all information appearing in or on the original labels, including the copyright and trademark notices are copied onto the labels of the copies.
+
+USES NOT PERMITTED:  You may not distribute, copy or use the Software except as explicitly permitted herein. Licensee has not been granted any trademark license as part of this Agreement and may not use the name or mark “OpenPose", "Carnegie Mellon" or any renditions thereof without the prior written permission of Licensor.
+
+You may not sell, rent, lease, sublicense, lend, time-share or transfer, in whole or in part, or provide third parties access to prior or present versions (or any parts thereof) of the Software.
+
+ASSIGNMENT: You may not assign this Agreement or your rights hereunder without the prior written consent of Licensor. Any attempted assignment without such consent shall be null and void.
+
+TERM: The term of the license granted by this Agreement is from Licensee's acceptance of this Agreement by downloading the Software or by using the Software until terminated as provided below.
+
+The Agreement automatically terminates without notice if you fail to comply with any provision of this Agreement.  Licensee may terminate this Agreement by ceasing using the Software.  Upon any termination of this Agreement, Licensee will delete any and all copies of the Software. You agree that all provisions which operate to protect the proprietary rights of Licensor shall remain in force should breach occur and that the obligation of confidentiality described in this Agreement is binding in perpetuity and, as such, survives the term of the Agreement.
+
+FEE: Provided Licensee abides completely by the terms and conditions of this Agreement, there is no fee due to Licensor for Licensee's use of the Software in accordance with this Agreement.
+
+DISCLAIMER OF WARRANTIES:  THE SOFTWARE IS PROVIDED "AS-IS" WITHOUT WARRANTY OF ANY KIND INCLUDING ANY WARRANTIES OF PERFORMANCE OR MERCHANTABILITY OR FITNESS FOR A PARTICULAR USE OR PURPOSE OR OF NON-INFRINGEMENT.  LICENSEE BEARS ALL RISK RELATING TO QUALITY AND PERFORMANCE OF THE SOFTWARE AND RELATED MATERIALS.
+
+SUPPORT AND MAINTENANCE: No Software support or training by the Licensor is provided as part of this Agreement.  
+
+EXCLUSIVE REMEDY AND LIMITATION OF LIABILITY: To the maximum extent permitted under applicable law, Licensor shall not be liable for direct, indirect, special, incidental, or consequential damages or lost profits related to Licensee's use of and/or inability to use the Software, even if Licensor is advised of the possibility of such damage.
+
+EXPORT REGULATION: Licensee agrees to comply with any and all applicable 
+U.S. export control laws, regulations, and/or other laws related to embargoes and sanction programs administered by the Office of Foreign Assets Control.
+
+SEVERABILITY: If any provision(s) of this Agreement shall be held to be invalid, illegal, or unenforceable by a court or other tribunal of competent jurisdiction, the validity, legality and enforceability of the remaining provisions shall not in any way be affected or impaired thereby.
+
+NO IMPLIED WAIVERS: No failure or delay by Licensor in enforcing any right or remedy under this Agreement shall be construed as a waiver of any future or other exercise of such right or remedy by Licensor.
+
+GOVERNING LAW: This Agreement shall be construed and enforced in accordance with the laws of the Commonwealth of Pennsylvania without reference to conflict of laws principles.  You consent to the personal jurisdiction of the courts of this County and waive their rights to venue outside of Allegheny County, Pennsylvania.
+
+ENTIRE AGREEMENT AND AMENDMENTS: This Agreement constitutes the sole and entire agreement between Licensee and Licensor as to the matter set forth herein and supersedes any previous agreements, understandings, and arrangements between the parties relating hereto.
+
+
+
+************************************************************************
+
+THIRD-PARTY SOFTWARE NOTICES AND INFORMATION
+
+This project incorporates material from the project(s) listed below (collectively, "Third Party Code").  This Third Party Code is licensed to you under their original license terms set forth below.  We reserves all other rights not expressly granted, whether by implication, estoppel or otherwise.
+ 
+1.	Caffe, version 1.0.0, (https://github.com/BVLC/caffe/)
+
+COPYRIGHT
+
+All contributions by the University of California:
+Copyright (c) 2014-2017 The Regents of the University of California (Regents)
+All rights reserved.
+
+All other contributions:
+Copyright (c) 2014-2017, the respective contributors
+All rights reserved.
+
+Caffe uses a shared copyright model: each contributor holds copyright over
+their contributions to Caffe. The project versioning records all such
+contribution and copyright details. If a contributor wants to further mark
+their specific copyright on a particular contribution, they should indicate
+their copyright solely in the commit message of the change when it is
+committed.
+
+LICENSE
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met: 
+
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer. 
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution. 
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+CONTRIBUTION AGREEMENT
+
+By contributing to the BVLC/caffe repository through pull-request, comment,
+or otherwise, the contributor releases their content to the
+license and copyright terms herein.
+
+************END OF THIRD-PARTY SOFTWARE NOTICES AND INFORMATION**********

negi/repos/controlnet_aux/src/controlnet_aux/open_pose/__init__.py

@@ -0,0 +1,234 @@
+# Openpose
+# Original from CMU https://github.com/CMU-Perceptual-Computing-Lab/openpose
+# 2nd Edited by https://github.com/Hzzone/pytorch-openpose
+# 3rd Edited by ControlNet
+# 4th Edited by ControlNet (added face and correct hands)
+# 5th Edited by ControlNet (Improved JSON serialization/deserialization, and lots of bug fixs)
+# This preprocessor is licensed by CMU for non-commercial use only.
+
+
+import os
+
+os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
+
+import json
+import warnings
+from typing import Callable, List, NamedTuple, Tuple, Union
+
+import cv2
+import numpy as np
+import torch
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+from ..util import HWC3, resize_image
+from . import util
+from .body import Body, BodyResult, Keypoint
+from .face import Face
+from .hand import Hand
+
+HandResult = List[Keypoint]
+FaceResult = List[Keypoint]
+
+class PoseResult(NamedTuple):
+    body: BodyResult
+    left_hand: Union[HandResult, None]
+    right_hand: Union[HandResult, None]
+    face: Union[FaceResult, None]
+
+def draw_poses(poses: List[PoseResult], H, W, draw_body=True, draw_hand=True, draw_face=True):
+    """
+    Draw the detected poses on an empty canvas.
+
+    Args:
+        poses (List[PoseResult]): A list of PoseResult objects containing the detected poses.
+        H (int): The height of the canvas.
+        W (int): The width of the canvas.
+        draw_body (bool, optional): Whether to draw body keypoints. Defaults to True.
+        draw_hand (bool, optional): Whether to draw hand keypoints. Defaults to True.
+        draw_face (bool, optional): Whether to draw face keypoints. Defaults to True.
+
+    Returns:
+        numpy.ndarray: A 3D numpy array representing the canvas with the drawn poses.
+    """
+    canvas = np.zeros(shape=(H, W, 3), dtype=np.uint8)
+
+    for pose in poses:
+        if draw_body:
+            canvas = util.draw_bodypose(canvas, pose.body.keypoints)
+
+        if draw_hand:
+            canvas = util.draw_handpose(canvas, pose.left_hand)
+            canvas = util.draw_handpose(canvas, pose.right_hand)
+
+        if draw_face:
+            canvas = util.draw_facepose(canvas, pose.face)
+
+    return canvas
+    
+    
+class OpenposeDetector:
+    """
+    A class for detecting human poses in images using the Openpose model.
+
+    Attributes:
+        model_dir (str): Path to the directory where the pose models are stored.
+    """
+    def __init__(self, body_estimation, hand_estimation=None, face_estimation=None):
+        self.body_estimation = body_estimation
+        self.hand_estimation = hand_estimation
+        self.face_estimation = face_estimation
+
+    @classmethod
+    def from_pretrained(cls, pretrained_model_or_path, filename=None, hand_filename=None, face_filename=None, cache_dir=None):
+
+        if pretrained_model_or_path == "lllyasviel/ControlNet":
+            filename = filename or "annotator/ckpts/body_pose_model.pth"
+            hand_filename = hand_filename or "annotator/ckpts/hand_pose_model.pth"
+            face_filename = face_filename or "facenet.pth"
+
+            face_pretrained_model_or_path = "lllyasviel/Annotators"
+        else:
+            filename = filename or "body_pose_model.pth"
+            hand_filename = hand_filename or "hand_pose_model.pth"
+            face_filename = face_filename or "facenet.pth"
+
+            face_pretrained_model_or_path = pretrained_model_or_path
+
+        if os.path.isdir(pretrained_model_or_path):
+            body_model_path = os.path.join(pretrained_model_or_path, filename)
+            hand_model_path = os.path.join(pretrained_model_or_path, hand_filename)
+            face_model_path = os.path.join(face_pretrained_model_or_path, face_filename)
+        else:
+            body_model_path = hf_hub_download(pretrained_model_or_path, filename, cache_dir=cache_dir)
+            hand_model_path = hf_hub_download(pretrained_model_or_path, hand_filename, cache_dir=cache_dir)
+            face_model_path = hf_hub_download(face_pretrained_model_or_path, face_filename, cache_dir=cache_dir)
+
+        body_estimation = Body(body_model_path)
+        hand_estimation = Hand(hand_model_path)
+        face_estimation = Face(face_model_path)
+
+        return cls(body_estimation, hand_estimation, face_estimation)
+
+    def to(self, device):
+        self.body_estimation.to(device)
+        self.hand_estimation.to(device)
+        self.face_estimation.to(device)
+        return self
+
+    def detect_hands(self, body: BodyResult, oriImg) -> Tuple[Union[HandResult, None], Union[HandResult, None]]:
+        left_hand = None
+        right_hand = None
+        H, W, _ = oriImg.shape
+        for x, y, w, is_left in util.handDetect(body, oriImg):
+            peaks = self.hand_estimation(oriImg[y:y+w, x:x+w, :]).astype(np.float32)
+            if peaks.ndim == 2 and peaks.shape[1] == 2:
+                peaks[:, 0] = np.where(peaks[:, 0] < 1e-6, -1, peaks[:, 0] + x) / float(W)
+                peaks[:, 1] = np.where(peaks[:, 1] < 1e-6, -1, peaks[:, 1] + y) / float(H)
+                
+                hand_result = [
+                    Keypoint(x=peak[0], y=peak[1])
+                    for peak in peaks
+                ]
+
+                if is_left:
+                    left_hand = hand_result
+                else:
+                    right_hand = hand_result
+
+        return left_hand, right_hand
+
+    def detect_face(self, body: BodyResult, oriImg) -> Union[FaceResult, None]:
+        face = util.faceDetect(body, oriImg)
+        if face is None:
+            return None
+        
+        x, y, w = face
+        H, W, _ = oriImg.shape
+        heatmaps = self.face_estimation(oriImg[y:y+w, x:x+w, :])
+        peaks = self.face_estimation.compute_peaks_from_heatmaps(heatmaps).astype(np.float32)
+        if peaks.ndim == 2 and peaks.shape[1] == 2:
+            peaks[:, 0] = np.where(peaks[:, 0] < 1e-6, -1, peaks[:, 0] + x) / float(W)
+            peaks[:, 1] = np.where(peaks[:, 1] < 1e-6, -1, peaks[:, 1] + y) / float(H)
+            return [
+                Keypoint(x=peak[0], y=peak[1])
+                for peak in peaks
+            ]
+        
+        return None
+
+    def detect_poses(self, oriImg, include_hand=False, include_face=False) -> List[PoseResult]:
+        """
+        Detect poses in the given image.
+            Args:
+                oriImg (numpy.ndarray): The input image for pose detection.
+                include_hand (bool, optional): Whether to include hand detection. Defaults to False.
+                include_face (bool, optional): Whether to include face detection. Defaults to False.
+
+        Returns:
+            List[PoseResult]: A list of PoseResult objects containing the detected poses.
+        """
+        oriImg = oriImg[:, :, ::-1].copy()
+        H, W, C = oriImg.shape
+        with torch.no_grad():
+            candidate, subset = self.body_estimation(oriImg)
+            bodies = self.body_estimation.format_body_result(candidate, subset)
+
+            results = []
+            for body in bodies:
+                left_hand, right_hand, face = (None,) * 3
+                if include_hand:
+                    left_hand, right_hand = self.detect_hands(body, oriImg)
+                if include_face:
+                    face = self.detect_face(body, oriImg)
+                
+                results.append(PoseResult(BodyResult(
+                    keypoints=[
+                        Keypoint(
+                            x=keypoint.x / float(W),
+                            y=keypoint.y / float(H)
+                        ) if keypoint is not None else None
+                        for keypoint in body.keypoints
+                    ], 
+                    total_score=body.total_score,
+                    total_parts=body.total_parts
+                ), left_hand, right_hand, face))
+            
+            return results
+        
+    def __call__(self, input_image, detect_resolution=512, image_resolution=512, include_body=True, include_hand=False, include_face=False, hand_and_face=None, output_type="pil", **kwargs):
+        if hand_and_face is not None:
+            warnings.warn("hand_and_face is deprecated. Use include_hand and include_face instead.", DeprecationWarning)
+            include_hand = hand_and_face
+            include_face = hand_and_face
+
+        if "return_pil" in kwargs:
+            warnings.warn("return_pil is deprecated. Use output_type instead.", DeprecationWarning)
+            output_type = "pil" if kwargs["return_pil"] else "np"
+        if type(output_type) is bool:
+            warnings.warn("Passing `True` or `False` to `output_type` is deprecated and will raise an error in future versions")
+            if output_type:
+                output_type = "pil"
+
+        if not isinstance(input_image, np.ndarray):
+            input_image = np.array(input_image, dtype=np.uint8)
+
+        input_image = HWC3(input_image)
+        input_image = resize_image(input_image, detect_resolution)
+        H, W, C = input_image.shape
+        
+        poses = self.detect_poses(input_image, include_hand, include_face)
+        canvas = draw_poses(poses, H, W, draw_body=include_body, draw_hand=include_hand, draw_face=include_face) 
+
+        detected_map = canvas
+        detected_map = HWC3(detected_map)
+        
+        img = resize_image(input_image, image_resolution)
+        H, W, C = img.shape
+
+        detected_map = cv2.resize(detected_map, (W, H), interpolation=cv2.INTER_LINEAR)
+
+        if output_type == "pil":
+            detected_map = Image.fromarray(detected_map)
+
+        return detected_map

negi/repos/controlnet_aux/src/controlnet_aux/open_pose/body.py

@@ -0,0 +1,260 @@
+import math
+from typing import List, NamedTuple, Union
+
+import cv2
+import numpy as np
+import torch
+from scipy.ndimage.filters import gaussian_filter
+
+from . import util
+from .model import bodypose_model
+
+
+class Keypoint(NamedTuple):
+    x: float
+    y: float
+    score: float = 1.0
+    id: int = -1
+
+
+class BodyResult(NamedTuple):
+    # Note: Using `Union` instead of `|` operator as the ladder is a Python
+    # 3.10 feature.
+    # Annotator code should be Python 3.8 Compatible, as controlnet repo uses
+    # Python 3.8 environment.
+    # https://github.com/lllyasviel/ControlNet/blob/d3284fcd0972c510635a4f5abe2eeb71dc0de524/environment.yaml#L6
+    keypoints: List[Union[Keypoint, None]]
+    total_score: float
+    total_parts: int
+
+
+class Body(object):
+    def __init__(self, model_path):
+        self.model = bodypose_model()
+        model_dict = util.transfer(self.model, torch.load(model_path))
+        self.model.load_state_dict(model_dict)
+        self.model.eval()
+
+    def to(self, device):
+        self.model.to(device)
+        return self
+
+    def __call__(self, oriImg):
+        device = next(iter(self.model.parameters())).device
+        # scale_search = [0.5, 1.0, 1.5, 2.0]
+        scale_search = [0.5]
+        boxsize = 368
+        stride = 8
+        padValue = 128
+        thre1 = 0.1
+        thre2 = 0.05
+        multiplier = [x * boxsize / oriImg.shape[0] for x in scale_search]
+        heatmap_avg = np.zeros((oriImg.shape[0], oriImg.shape[1], 19))
+        paf_avg = np.zeros((oriImg.shape[0], oriImg.shape[1], 38))
+
+        for m in range(len(multiplier)):
+            scale = multiplier[m]
+            imageToTest = util.smart_resize_k(oriImg, fx=scale, fy=scale)
+            imageToTest_padded, pad = util.padRightDownCorner(imageToTest, stride, padValue)
+            im = np.transpose(np.float32(imageToTest_padded[:, :, :, np.newaxis]), (3, 2, 0, 1)) / 256 - 0.5
+            im = np.ascontiguousarray(im)
+
+            data = torch.from_numpy(im).float()
+            data = data.to(device)
+            # data = data.permute([2, 0, 1]).unsqueeze(0).float()
+            with torch.no_grad():
+                Mconv7_stage6_L1, Mconv7_stage6_L2 = self.model(data)
+            Mconv7_stage6_L1 = Mconv7_stage6_L1.cpu().numpy()
+            Mconv7_stage6_L2 = Mconv7_stage6_L2.cpu().numpy()
+
+            # extract outputs, resize, and remove padding
+            # heatmap = np.transpose(np.squeeze(net.blobs[output_blobs.keys()[1]].data), (1, 2, 0))  # output 1 is heatmaps
+            heatmap = np.transpose(np.squeeze(Mconv7_stage6_L2), (1, 2, 0))  # output 1 is heatmaps
+            heatmap = util.smart_resize_k(heatmap, fx=stride, fy=stride)
+            heatmap = heatmap[:imageToTest_padded.shape[0] - pad[2], :imageToTest_padded.shape[1] - pad[3], :]
+            heatmap = util.smart_resize(heatmap, (oriImg.shape[0], oriImg.shape[1]))
+
+            # paf = np.transpose(np.squeeze(net.blobs[output_blobs.keys()[0]].data), (1, 2, 0))  # output 0 is PAFs
+            paf = np.transpose(np.squeeze(Mconv7_stage6_L1), (1, 2, 0))  # output 0 is PAFs
+            paf = util.smart_resize_k(paf, fx=stride, fy=stride)
+            paf = paf[:imageToTest_padded.shape[0] - pad[2], :imageToTest_padded.shape[1] - pad[3], :]
+            paf = util.smart_resize(paf, (oriImg.shape[0], oriImg.shape[1]))
+
+            heatmap_avg += heatmap_avg + heatmap / len(multiplier)
+            paf_avg += + paf / len(multiplier)
+
+        all_peaks = []
+        peak_counter = 0
+
+        for part in range(18):
+            map_ori = heatmap_avg[:, :, part]
+            one_heatmap = gaussian_filter(map_ori, sigma=3)
+
+            map_left = np.zeros(one_heatmap.shape)
+            map_left[1:, :] = one_heatmap[:-1, :]
+            map_right = np.zeros(one_heatmap.shape)
+            map_right[:-1, :] = one_heatmap[1:, :]
+            map_up = np.zeros(one_heatmap.shape)
+            map_up[:, 1:] = one_heatmap[:, :-1]
+            map_down = np.zeros(one_heatmap.shape)
+            map_down[:, :-1] = one_heatmap[:, 1:]
+
+            peaks_binary = np.logical_and.reduce(
+                (one_heatmap >= map_left, one_heatmap >= map_right, one_heatmap >= map_up, one_heatmap >= map_down, one_heatmap > thre1))
+            peaks = list(zip(np.nonzero(peaks_binary)[1], np.nonzero(peaks_binary)[0]))  # note reverse
+            peaks_with_score = [x + (map_ori[x[1], x[0]],) for x in peaks]
+            peak_id = range(peak_counter, peak_counter + len(peaks))
+            peaks_with_score_and_id = [peaks_with_score[i] + (peak_id[i],) for i in range(len(peak_id))]
+
+            all_peaks.append(peaks_with_score_and_id)
+            peak_counter += len(peaks)
+
+        # find connection in the specified sequence, center 29 is in the position 15
+        limbSeq = [[2, 3], [2, 6], [3, 4], [4, 5], [6, 7], [7, 8], [2, 9], [9, 10], \
+                   [10, 11], [2, 12], [12, 13], [13, 14], [2, 1], [1, 15], [15, 17], \
+                   [1, 16], [16, 18], [3, 17], [6, 18]]
+        # the middle joints heatmap correpondence
+        mapIdx = [[31, 32], [39, 40], [33, 34], [35, 36], [41, 42], [43, 44], [19, 20], [21, 22], \
+                  [23, 24], [25, 26], [27, 28], [29, 30], [47, 48], [49, 50], [53, 54], [51, 52], \
+                  [55, 56], [37, 38], [45, 46]]
+
+        connection_all = []
+        special_k = []
+        mid_num = 10
+
+        for k in range(len(mapIdx)):
+            score_mid = paf_avg[:, :, [x - 19 for x in mapIdx[k]]]
+            candA = all_peaks[limbSeq[k][0] - 1]
+            candB = all_peaks[limbSeq[k][1] - 1]
+            nA = len(candA)
+            nB = len(candB)
+            indexA, indexB = limbSeq[k]
+            if (nA != 0 and nB != 0):
+                connection_candidate = []
+                for i in range(nA):
+                    for j in range(nB):
+                        vec = np.subtract(candB[j][:2], candA[i][:2])
+                        norm = math.sqrt(vec[0] * vec[0] + vec[1] * vec[1])
+                        norm = max(0.001, norm)
+                        vec = np.divide(vec, norm)
+
+                        startend = list(zip(np.linspace(candA[i][0], candB[j][0], num=mid_num), \
+                                            np.linspace(candA[i][1], candB[j][1], num=mid_num)))
+
+                        vec_x = np.array([score_mid[int(round(startend[I][1])), int(round(startend[I][0])), 0] \
+                                          for I in range(len(startend))])
+                        vec_y = np.array([score_mid[int(round(startend[I][1])), int(round(startend[I][0])), 1] \
+                                          for I in range(len(startend))])
+
+                        score_midpts = np.multiply(vec_x, vec[0]) + np.multiply(vec_y, vec[1])
+                        score_with_dist_prior = sum(score_midpts) / len(score_midpts) + min(
+                            0.5 * oriImg.shape[0] / norm - 1, 0)
+                        criterion1 = len(np.nonzero(score_midpts > thre2)[0]) > 0.8 * len(score_midpts)
+                        criterion2 = score_with_dist_prior > 0
+                        if criterion1 and criterion2:
+                            connection_candidate.append(
+                                [i, j, score_with_dist_prior, score_with_dist_prior + candA[i][2] + candB[j][2]])
+
+                connection_candidate = sorted(connection_candidate, key=lambda x: x[2], reverse=True)
+                connection = np.zeros((0, 5))
+                for c in range(len(connection_candidate)):
+                    i, j, s = connection_candidate[c][0:3]
+                    if (i not in connection[:, 3] and j not in connection[:, 4]):
+                        connection = np.vstack([connection, [candA[i][3], candB[j][3], s, i, j]])
+                        if (len(connection) >= min(nA, nB)):
+                            break
+
+                connection_all.append(connection)
+            else:
+                special_k.append(k)
+                connection_all.append([])
+
+        # last number in each row is the total parts number of that person
+        # the second last number in each row is the score of the overall configuration
+        subset = -1 * np.ones((0, 20))
+        candidate = np.array([item for sublist in all_peaks for item in sublist])
+
+        for k in range(len(mapIdx)):
+            if k not in special_k:
+                partAs = connection_all[k][:, 0]
+                partBs = connection_all[k][:, 1]
+                indexA, indexB = np.array(limbSeq[k]) - 1
+
+                for i in range(len(connection_all[k])):  # = 1:size(temp,1)
+                    found = 0
+                    subset_idx = [-1, -1]
+                    for j in range(len(subset)):  # 1:size(subset,1):
+                        if subset[j][indexA] == partAs[i] or subset[j][indexB] == partBs[i]:
+                            subset_idx[found] = j
+                            found += 1
+
+                    if found == 1:
+                        j = subset_idx[0]
+                        if subset[j][indexB] != partBs[i]:
+                            subset[j][indexB] = partBs[i]
+                            subset[j][-1] += 1
+                            subset[j][-2] += candidate[partBs[i].astype(int), 2] + connection_all[k][i][2]
+                    elif found == 2:  # if found 2 and disjoint, merge them
+                        j1, j2 = subset_idx
+                        membership = ((subset[j1] >= 0).astype(int) + (subset[j2] >= 0).astype(int))[:-2]
+                        if len(np.nonzero(membership == 2)[0]) == 0:  # merge
+                            subset[j1][:-2] += (subset[j2][:-2] + 1)
+                            subset[j1][-2:] += subset[j2][-2:]
+                            subset[j1][-2] += connection_all[k][i][2]
+                            subset = np.delete(subset, j2, 0)
+                        else:  # as like found == 1
+                            subset[j1][indexB] = partBs[i]
+                            subset[j1][-1] += 1
+                            subset[j1][-2] += candidate[partBs[i].astype(int), 2] + connection_all[k][i][2]
+
+                    # if find no partA in the subset, create a new subset
+                    elif not found and k < 17:
+                        row = -1 * np.ones(20)
+                        row[indexA] = partAs[i]
+                        row[indexB] = partBs[i]
+                        row[-1] = 2
+                        row[-2] = sum(candidate[connection_all[k][i, :2].astype(int), 2]) + connection_all[k][i][2]
+                        subset = np.vstack([subset, row])
+        # delete some rows of subset which has few parts occur
+        deleteIdx = []
+        for i in range(len(subset)):
+            if subset[i][-1] < 4 or subset[i][-2] / subset[i][-1] < 0.4:
+                deleteIdx.append(i)
+        subset = np.delete(subset, deleteIdx, axis=0)
+
+        # subset: n*20 array, 0-17 is the index in candidate, 18 is the total score, 19 is the total parts
+        # candidate: x, y, score, id
+        return candidate, subset
+    
+    @staticmethod
+    def format_body_result(candidate: np.ndarray, subset: np.ndarray) -> List[BodyResult]:
+        """
+        Format the body results from the candidate and subset arrays into a list of BodyResult objects.
+        
+        Args:
+            candidate (np.ndarray): An array of candidates containing the x, y coordinates, score, and id
+                for each body part.
+            subset (np.ndarray): An array of subsets containing indices to the candidate array for each
+                person detected. The last two columns of each row hold the total score and total parts
+                of the person.
+
+        Returns:
+            List[BodyResult]: A list of BodyResult objects, where each object represents a person with
+                detected keypoints, total score, and total parts.
+        """
+        return [
+            BodyResult(
+                keypoints=[
+                    Keypoint(
+                        x=candidate[candidate_index][0],
+                        y=candidate[candidate_index][1],
+                        score=candidate[candidate_index][2],
+                        id=candidate[candidate_index][3]
+                    ) if candidate_index != -1 else None
+                    for candidate_index in person[:18].astype(int)
+                ],
+                total_score=person[18],
+                total_parts=person[19]
+            )
+            for person in subset
+        ]

negi/repos/controlnet_aux/src/controlnet_aux/open_pose/face.py

@@ -0,0 +1,364 @@
+import logging
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+from torch.nn import Conv2d, MaxPool2d, Module, ReLU, init
+from torchvision.transforms import ToPILImage, ToTensor
+
+from . import util
+
+
+class FaceNet(Module):
+    """Model the cascading heatmaps. """
+    def __init__(self):
+        super(FaceNet, self).__init__()
+        # cnn to make feature map
+        self.relu = ReLU()
+        self.max_pooling_2d = MaxPool2d(kernel_size=2, stride=2)
+        self.conv1_1 = Conv2d(in_channels=3, out_channels=64,
+                              kernel_size=3, stride=1, padding=1)
+        self.conv1_2 = Conv2d(
+            in_channels=64, out_channels=64, kernel_size=3, stride=1,
+            padding=1)
+        self.conv2_1 = Conv2d(
+            in_channels=64, out_channels=128, kernel_size=3, stride=1,
+            padding=1)
+        self.conv2_2 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=3, stride=1,
+            padding=1)
+        self.conv3_1 = Conv2d(
+            in_channels=128, out_channels=256, kernel_size=3, stride=1,
+            padding=1)
+        self.conv3_2 = Conv2d(
+            in_channels=256, out_channels=256, kernel_size=3, stride=1,
+            padding=1)
+        self.conv3_3 = Conv2d(
+            in_channels=256, out_channels=256, kernel_size=3, stride=1,
+            padding=1)
+        self.conv3_4 = Conv2d(
+            in_channels=256, out_channels=256, kernel_size=3, stride=1,
+            padding=1)
+        self.conv4_1 = Conv2d(
+            in_channels=256, out_channels=512, kernel_size=3, stride=1,
+            padding=1)
+        self.conv4_2 = Conv2d(
+            in_channels=512, out_channels=512, kernel_size=3, stride=1,
+            padding=1)
+        self.conv4_3 = Conv2d(
+            in_channels=512, out_channels=512, kernel_size=3, stride=1,
+            padding=1)
+        self.conv4_4 = Conv2d(
+            in_channels=512, out_channels=512, kernel_size=3, stride=1,
+            padding=1)
+        self.conv5_1 = Conv2d(
+            in_channels=512, out_channels=512, kernel_size=3, stride=1,
+            padding=1)
+        self.conv5_2 = Conv2d(
+            in_channels=512, out_channels=512, kernel_size=3, stride=1,
+            padding=1)
+        self.conv5_3_CPM = Conv2d(
+            in_channels=512, out_channels=128, kernel_size=3, stride=1,
+            padding=1)
+
+        # stage1
+        self.conv6_1_CPM = Conv2d(
+            in_channels=128, out_channels=512, kernel_size=1, stride=1,
+            padding=0)
+        self.conv6_2_CPM = Conv2d(
+            in_channels=512, out_channels=71, kernel_size=1, stride=1,
+            padding=0)
+
+        # stage2
+        self.Mconv1_stage2 = Conv2d(
+            in_channels=199, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv2_stage2 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv3_stage2 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv4_stage2 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv5_stage2 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv6_stage2 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=1, stride=1,
+            padding=0)
+        self.Mconv7_stage2 = Conv2d(
+            in_channels=128, out_channels=71, kernel_size=1, stride=1,
+            padding=0)
+
+        # stage3
+        self.Mconv1_stage3 = Conv2d(
+            in_channels=199, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv2_stage3 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv3_stage3 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv4_stage3 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv5_stage3 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv6_stage3 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=1, stride=1,
+            padding=0)
+        self.Mconv7_stage3 = Conv2d(
+            in_channels=128, out_channels=71, kernel_size=1, stride=1,
+            padding=0)
+
+        # stage4
+        self.Mconv1_stage4 = Conv2d(
+            in_channels=199, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv2_stage4 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv3_stage4 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv4_stage4 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv5_stage4 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv6_stage4 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=1, stride=1,
+            padding=0)
+        self.Mconv7_stage4 = Conv2d(
+            in_channels=128, out_channels=71, kernel_size=1, stride=1,
+            padding=0)
+
+        # stage5
+        self.Mconv1_stage5 = Conv2d(
+            in_channels=199, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv2_stage5 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv3_stage5 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv4_stage5 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv5_stage5 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv6_stage5 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=1, stride=1,
+            padding=0)
+        self.Mconv7_stage5 = Conv2d(
+            in_channels=128, out_channels=71, kernel_size=1, stride=1,
+            padding=0)
+
+        # stage6
+        self.Mconv1_stage6 = Conv2d(
+            in_channels=199, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv2_stage6 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv3_stage6 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv4_stage6 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv5_stage6 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=7, stride=1,
+            padding=3)
+        self.Mconv6_stage6 = Conv2d(
+            in_channels=128, out_channels=128, kernel_size=1, stride=1,
+            padding=0)
+        self.Mconv7_stage6 = Conv2d(
+            in_channels=128, out_channels=71, kernel_size=1, stride=1,
+            padding=0)
+
+        for m in self.modules():
+            if isinstance(m, Conv2d):
+                init.constant_(m.bias, 0)
+
+    def forward(self, x):
+        """Return a list of heatmaps."""
+        heatmaps = []
+
+        h = self.relu(self.conv1_1(x))
+        h = self.relu(self.conv1_2(h))
+        h = self.max_pooling_2d(h)
+        h = self.relu(self.conv2_1(h))
+        h = self.relu(self.conv2_2(h))
+        h = self.max_pooling_2d(h)
+        h = self.relu(self.conv3_1(h))
+        h = self.relu(self.conv3_2(h))
+        h = self.relu(self.conv3_3(h))
+        h = self.relu(self.conv3_4(h))
+        h = self.max_pooling_2d(h)
+        h = self.relu(self.conv4_1(h))
+        h = self.relu(self.conv4_2(h))
+        h = self.relu(self.conv4_3(h))
+        h = self.relu(self.conv4_4(h))
+        h = self.relu(self.conv5_1(h))
+        h = self.relu(self.conv5_2(h))
+        h = self.relu(self.conv5_3_CPM(h))
+        feature_map = h
+
+        # stage1
+        h = self.relu(self.conv6_1_CPM(h))
+        h = self.conv6_2_CPM(h)
+        heatmaps.append(h)
+
+        # stage2
+        h = torch.cat([h, feature_map], dim=1)  # channel concat
+        h = self.relu(self.Mconv1_stage2(h))
+        h = self.relu(self.Mconv2_stage2(h))
+        h = self.relu(self.Mconv3_stage2(h))
+        h = self.relu(self.Mconv4_stage2(h))
+        h = self.relu(self.Mconv5_stage2(h))
+        h = self.relu(self.Mconv6_stage2(h))
+        h = self.Mconv7_stage2(h)
+        heatmaps.append(h)
+
+        # stage3
+        h = torch.cat([h, feature_map], dim=1)  # channel concat
+        h = self.relu(self.Mconv1_stage3(h))
+        h = self.relu(self.Mconv2_stage3(h))
+        h = self.relu(self.Mconv3_stage3(h))
+        h = self.relu(self.Mconv4_stage3(h))
+        h = self.relu(self.Mconv5_stage3(h))
+        h = self.relu(self.Mconv6_stage3(h))
+        h = self.Mconv7_stage3(h)
+        heatmaps.append(h)
+
+        # stage4
+        h = torch.cat([h, feature_map], dim=1)  # channel concat
+        h = self.relu(self.Mconv1_stage4(h))
+        h = self.relu(self.Mconv2_stage4(h))
+        h = self.relu(self.Mconv3_stage4(h))
+        h = self.relu(self.Mconv4_stage4(h))
+        h = self.relu(self.Mconv5_stage4(h))
+        h = self.relu(self.Mconv6_stage4(h))
+        h = self.Mconv7_stage4(h)
+        heatmaps.append(h)
+
+        # stage5
+        h = torch.cat([h, feature_map], dim=1)  # channel concat
+        h = self.relu(self.Mconv1_stage5(h))
+        h = self.relu(self.Mconv2_stage5(h))
+        h = self.relu(self.Mconv3_stage5(h))
+        h = self.relu(self.Mconv4_stage5(h))
+        h = self.relu(self.Mconv5_stage5(h))
+        h = self.relu(self.Mconv6_stage5(h))
+        h = self.Mconv7_stage5(h)
+        heatmaps.append(h)
+
+        # stage6
+        h = torch.cat([h, feature_map], dim=1)  # channel concat
+        h = self.relu(self.Mconv1_stage6(h))
+        h = self.relu(self.Mconv2_stage6(h))
+        h = self.relu(self.Mconv3_stage6(h))
+        h = self.relu(self.Mconv4_stage6(h))
+        h = self.relu(self.Mconv5_stage6(h))
+        h = self.relu(self.Mconv6_stage6(h))
+        h = self.Mconv7_stage6(h)
+        heatmaps.append(h)
+
+        return heatmaps
+
+
+LOG = logging.getLogger(__name__)
+TOTEN = ToTensor()
+TOPIL = ToPILImage()
+
+
+params = {
+    'gaussian_sigma': 2.5,
+    'inference_img_size': 736,  # 368, 736, 1312
+    'heatmap_peak_thresh': 0.1,
+    'crop_scale': 1.5,
+    'line_indices': [
+        [0, 1], [1, 2], [2, 3], [3, 4], [4, 5], [5, 6],
+        [6, 7], [7, 8], [8, 9], [9, 10], [10, 11], [11, 12], [12, 13],
+        [13, 14], [14, 15], [15, 16],
+        [17, 18], [18, 19], [19, 20], [20, 21],
+        [22, 23], [23, 24], [24, 25], [25, 26],
+        [27, 28], [28, 29], [29, 30],
+        [31, 32], [32, 33], [33, 34], [34, 35],
+        [36, 37], [37, 38], [38, 39], [39, 40], [40, 41], [41, 36],
+        [42, 43], [43, 44], [44, 45], [45, 46], [46, 47], [47, 42],
+        [48, 49], [49, 50], [50, 51], [51, 52], [52, 53], [53, 54],
+        [54, 55], [55, 56], [56, 57], [57, 58], [58, 59], [59, 48],
+        [60, 61], [61, 62], [62, 63], [63, 64], [64, 65], [65, 66],
+        [66, 67], [67, 60]
+    ],
+}
+
+
+class Face(object):
+    """
+    The OpenPose face landmark detector model.
+
+    Args:
+        inference_size: set the size of the inference image size, suggested:
+            368, 736, 1312, default 736
+        gaussian_sigma: blur the heatmaps, default 2.5
+        heatmap_peak_thresh: return landmark if over threshold, default 0.1
+
+    """
+    def __init__(self, face_model_path,
+                 inference_size=None,
+                 gaussian_sigma=None,
+                 heatmap_peak_thresh=None):
+        self.inference_size = inference_size or params["inference_img_size"]
+        self.sigma = gaussian_sigma or params['gaussian_sigma']
+        self.threshold = heatmap_peak_thresh or params["heatmap_peak_thresh"]
+        self.model = FaceNet()
+        self.model.load_state_dict(torch.load(face_model_path))
+        self.model.eval()
+
+    def to(self, device):
+        self.model.to(device)
+        return self
+
+    def __call__(self, face_img):
+        device = next(iter(self.model.parameters())).device
+        H, W, C = face_img.shape
+
+        w_size = 384
+        x_data = torch.from_numpy(util.smart_resize(face_img, (w_size, w_size))).permute([2, 0, 1]) / 256.0 - 0.5
+
+        x_data = x_data.to(device)
+
+        with torch.no_grad():
+            hs = self.model(x_data[None, ...])
+            heatmaps = F.interpolate(
+                hs[-1],
+                (H, W),
+                mode='bilinear', align_corners=True).cpu().numpy()[0]
+        return heatmaps
+
+    def compute_peaks_from_heatmaps(self, heatmaps):
+        all_peaks = []
+        for part in range(heatmaps.shape[0]):
+            map_ori = heatmaps[part].copy()
+            binary = np.ascontiguousarray(map_ori > 0.05, dtype=np.uint8)
+
+            if np.sum(binary) == 0:
+                continue
+
+            positions = np.where(binary > 0.5)
+            intensities = map_ori[positions]
+            mi = np.argmax(intensities)
+            y, x = positions[0][mi], positions[1][mi]
+            all_peaks.append([x, y])
+
+        return np.array(all_peaks)