241 lines
9.2 KiB
Python
241 lines
9.2 KiB
Python
# !/usr/bin/env python
|
|
# -*- coding: UTF-8 -*-
|
|
import sys
|
|
import os
|
|
import numpy as np
|
|
import torch
|
|
from omegaconf import OmegaConf
|
|
from pytorch_lightning import seed_everything
|
|
import copy
|
|
from .custom_net.ddim import DDIMSampler
|
|
from einops import rearrange
|
|
import math
|
|
from PIL import Image, ImageDraw
|
|
from .custom_net.customnet_util import instantiate_from_config, img2tensor
|
|
from .gradio_utils import load_preprocess_model, preprocess_image
|
|
from comfy.utils import common_upscale
|
|
import folder_paths
|
|
from folder_paths import base_path
|
|
sys.path.append(os.path.join(base_path,"custom_nodes","ComfyUI_CustomNet","custom_net"))
|
|
cur_path = os.path.dirname(os.path.abspath(__file__))
|
|
|
|
def get_instance_path(path):
|
|
os_path = os.path.normpath(path)
|
|
if sys.platform.startswith('win32'):
|
|
os_path = os_path.replace('\\', "/")
|
|
return os_path
|
|
|
|
def tensor_to_image( tensor):
|
|
image_np = tensor.squeeze().mul(255).clamp(0, 255).byte().numpy()
|
|
image = Image.fromarray(image_np, mode='RGB')
|
|
return image
|
|
|
|
def upscale_to_pil(img_tensor, width, height):
|
|
samples = img_tensor.movedim(-1, 1)
|
|
img = common_upscale(samples, width, height, "nearest-exact", "center")
|
|
samples = img.movedim(1, -1)
|
|
img_pil = tensor_to_image(samples)
|
|
return img_pil
|
|
|
|
def preprocess_input(preprocess_model, input_image): # reg background
|
|
processed_image = preprocess_image(preprocess_model, input_image)
|
|
# input_img = (processed_image / 255.0).astype(np.float32)
|
|
return processed_image
|
|
# return processed_image, processed_image
|
|
|
|
|
|
def prepare_data(device, input_image, x0, y0, x1, y1, polar, azimuth, prompt,bg_image,use_inpaint):
|
|
|
|
# if input_image.size[0] != 256 or input_image.size[1] != 256:
|
|
# input_image = input_image.resize((256, 256))
|
|
# input_image = np.array(input_image)
|
|
# img_cond = img2tensor(input_image, bgr2rgb=False, float32=True).unsqueeze(0) / 255.
|
|
input_image = np.array(input_image)
|
|
img_cond = img2tensor(input_image, bgr2rgb=False, float32=True).unsqueeze(0) / 255.
|
|
img_cond = img_cond * 2 - 1
|
|
|
|
img_location = copy.deepcopy(img_cond)
|
|
input_im_padding = torch.ones_like(img_location)
|
|
|
|
x_0 = min(x0, x1)
|
|
x_1 = max(x0, x1)
|
|
y_0 = min(y0, y1)
|
|
y_1 = max(y0, y1)
|
|
|
|
# print(x0, y0, x1, y1)
|
|
# print(x_0, y_0, x_1, y_1)
|
|
img_location = torch.nn.functional.interpolate(img_location, (y_1 - y_0, x_1 - x_0), mode="bilinear")
|
|
input_im_padding[:, :, y_0:y_1, x_0:x_1] = img_location
|
|
img_location = input_im_padding
|
|
|
|
if use_inpaint:
|
|
bg_image = np.array(bg_image)
|
|
bg_cond=img2tensor( bg_image, bgr2rgb=False, float32=True) / 255.
|
|
bg_cond =bg_cond* 2 - 1
|
|
bg_cond =bg_cond.unsqueeze(0)
|
|
bg_cond[:, :, y_0:y_1, x_0:x_1] = 1
|
|
else:
|
|
bg_cond=img_cond #no use
|
|
|
|
T = torch.tensor(
|
|
[[math.radians(polar), math.sin(math.radians(azimuth)), math.cos(math.radians(azimuth)), 0.0]]).unsqueeze(1)
|
|
|
|
|
|
batch = {
|
|
"image_cond": img_cond.to(device),
|
|
"image_location": img_location.to(device),
|
|
"bg_cond":bg_cond.to(device),
|
|
'T': T.to(device),
|
|
'text': [prompt],
|
|
}
|
|
return batch
|
|
|
|
|
|
device = torch.device("cuda")
|
|
class CustomNet_LoadModel:
|
|
def __init__(self):
|
|
pass
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"ckpt_name": (["none"] + folder_paths.get_filename_list("checkpoints"),),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("MODEL","DICT",)
|
|
RETURN_NAMES = ("model","info")
|
|
FUNCTION = "main_loader"
|
|
CATEGORY = "CustomNet_Plus"
|
|
|
|
def main_loader(self,ckpt_name,):
|
|
# load model
|
|
ckpt = folder_paths.get_full_path("checkpoints", ckpt_name)
|
|
if "inpaint"in ckpt_name:
|
|
path_yaml = os.path.join(cur_path, "configs", "config_customnet_inpaint.yaml")
|
|
use_inpaint=True
|
|
else:
|
|
path_yaml = os.path.join(cur_path, "configs", "config_customnet.yaml")
|
|
use_inpaint = False
|
|
path_yaml = get_instance_path(path_yaml)
|
|
config = OmegaConf.load(path_yaml)
|
|
model = instantiate_from_config(config.model)
|
|
ckpt_load = torch.load(ckpt, map_location="cpu")
|
|
model.load_state_dict(ckpt_load, strict=False)
|
|
del ckpt_load
|
|
model = model.to(device)
|
|
info={"inpaint":use_inpaint,}
|
|
return (model,info,)
|
|
|
|
class CustomNet_Sampler:
|
|
|
|
def __init__(self):
|
|
pass
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"model":("MODEL",),
|
|
"info":("DICT",),
|
|
"image": ("IMAGE",),
|
|
"prompt": ("STRING", {"multiline": True, "default": "on the seaside"}),
|
|
"neg_prompt": ("STRING", {"multiline": True,"default": ""}),
|
|
"steps": ("INT", {"default": 50, "min": 1, "max": 1024,"step": 1,"display": "number"}),
|
|
"seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
|
|
"width": ("INT", {"default": 256, "min": 128, "max": 512, "step": 64, "display": "number"}),
|
|
"height": ("INT", {"default": 256, "min": 128, "max": 512, "step": 64, "display": "number"}),
|
|
"obj_x": ("INT", {"default": 50, "min": 0, "max": 256, "step": 1, "display": "number"}),
|
|
"obj_y": ("INT", {"default": 50, "min": 0, "max": 256, "step": 1, "display": "number"}),
|
|
"bg_x": ("INT", {"default": 200, "min": 0, "max": 256, "step": 1, "display": "number"}),
|
|
"bg_y": ("INT", {"default": 200, "min": 0, "max": 256, "step": 1, "display": "number"}),
|
|
"polar": ("FLOAT", {"default": 0, "min": -30, "max": 30, "step": -0.5, "display": "number"}),
|
|
"azimuth": ("FLOAT", {"default": 0, "min": -60, "max": 30, "step": -0.5, "display": "number"}),
|
|
"batch_size": ("INT", {"default": 1, "min": 1, "max": 256, "step": 1, "display": "number"}), },
|
|
"optional": {
|
|
"bg_image": ("IMAGE",), }
|
|
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE",)
|
|
RETURN_NAMES = ("output_image", )
|
|
FUNCTION = "customnet_main"
|
|
CATEGORY = "CustomNet_Plus"
|
|
|
|
|
|
@torch.no_grad()
|
|
def customnet_main(self,model,info,image, prompt,neg_prompt,steps,seed,width,height,obj_x, obj_y, bg_x, bg_y, polar, azimuth,batch_size,**kwargs):
|
|
preprocess_model = load_preprocess_model()
|
|
sampler = DDIMSampler(model, device=device)
|
|
use_inpaint=info["inpaint"]
|
|
print(f"inpaint is {use_inpaint}")
|
|
input_image = upscale_to_pil(image, width, height) # comfy upscale tensor2pil
|
|
if use_inpaint:
|
|
bg_image=kwargs.get("bg_image")
|
|
bg_image=upscale_to_pil(bg_image, width, height)
|
|
else:
|
|
bg_image=input_image
|
|
|
|
input_image = preprocess_input(preprocess_model, input_image) # using interface reg img
|
|
seed_everything(seed)
|
|
|
|
batch = prepare_data(device, input_image, obj_x, obj_y, bg_x, bg_y, polar, azimuth, prompt,bg_image,use_inpaint)
|
|
|
|
c = model.get_learned_conditioning(batch["image_cond"])
|
|
c = torch.cat([c, batch["T"]], dim=-1)
|
|
c = model.cc_projection(c)
|
|
if use_inpaint:
|
|
bg_concat = model.encode_first_stage(batch["bg_cond"]).mode().detach()
|
|
|
|
## condition
|
|
cond = {}
|
|
cond['c_concat'] = [model.encode_first_stage((batch["image_location"])).mode().detach()]
|
|
cond['c_crossattn'] = [c]
|
|
text_embedding = model.text_encoder(batch["text"])
|
|
cond["c_crossattn"].append(text_embedding)
|
|
if use_inpaint:
|
|
cond['c_concat'].append(bg_concat)
|
|
|
|
## null-condition
|
|
uc = {}
|
|
uc['c_concat'] = [torch.zeros(1, 4, 32, 32).to(c.device)]
|
|
uc['c_crossattn'] = [torch.zeros_like(c).to(c.device)]
|
|
uc_text_embedding = model.text_encoder([neg_prompt])
|
|
uc['c_crossattn'].append(uc_text_embedding)
|
|
if use_inpaint:
|
|
uc["c_concat"].append(bg_concat)
|
|
|
|
|
|
## sample
|
|
shape = [4, 32, 32]
|
|
samples_latents, _ = sampler.sample(
|
|
S=steps,
|
|
batch_size=batch_size,
|
|
shape=shape,
|
|
verbose=False,
|
|
unconditional_guidance_scale=999, # useless
|
|
conditioning=cond,
|
|
unconditional_conditioning=uc,
|
|
cfg_type=0,
|
|
cfg_scale_dict={"img": 0., "text": 0., "all": 3.0}
|
|
)
|
|
|
|
x_samples = model.decode_first_stage(samples_latents)
|
|
x_samples = torch.clamp((x_samples + 1.0) / 2.0, min=0.0, max=1.0).cpu().numpy()
|
|
x_samples = rearrange(255.0 * x_samples[0], 'c h w -> h w c').astype(np.uint8)
|
|
image = Image.fromarray(x_samples)
|
|
image = torch.from_numpy(np.array(image).astype(np.float32) / 255.0).unsqueeze(0)
|
|
return (image,)
|
|
|
|
|
|
NODE_CLASS_MAPPINGS = {
|
|
"CustomNet_LoadModel":CustomNet_LoadModel,
|
|
"CustomNet_Sampler": CustomNet_Sampler
|
|
}
|
|
|
|
NODE_DISPLAY_NAME_MAPPINGS = {
|
|
"CustomNet_LoadModel":"CustomNet_LoadModel",
|
|
"CustomNet_Sampler": "CustomNet_Sampler"
|
|
}
|