353 lines
12 KiB
Python
353 lines
12 KiB
Python
import os
|
|
|
|
import PIL
|
|
import cv2
|
|
import numpy as np
|
|
import torch
|
|
from PIL import Image, ImageDraw
|
|
from detectron2.checkpoint import DetectionCheckpointer
|
|
from detectron2.config import LazyConfig, instantiate
|
|
from groundingdino.config import GroundingDINO_SwinT_OGC
|
|
from groundingdino.util.inference import Model as DinoModel
|
|
from segment_anything import sam_model_registry, SamPredictor
|
|
import groundingdino.datasets.transforms as T
|
|
from torch import Tensor
|
|
|
|
import folder_paths
|
|
import comfy.model_management
|
|
|
|
folder_paths.folder_names_and_paths["sams"] = ([os.path.join(folder_paths.models_dir, "sams")],
|
|
folder_paths.supported_pt_extensions)
|
|
|
|
|
|
def to_dino_format(image):
|
|
dino_transform = T.Compose(
|
|
[
|
|
T.RandomResize([800], max_size=1333),
|
|
T.ToTensor(),
|
|
T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
|
|
]
|
|
)
|
|
image_transformed, _ = dino_transform(Image.fromarray(image), None)
|
|
return image_transformed
|
|
|
|
# https://github.com/WASasquatch/was-node-suite-comfyui/blob/4413ecd432aa1f13d8ad01a4f32fe8fe506e7051/WAS_Node_Suite.py#L374
|
|
# Tensor to SAM-compatible NumPy
|
|
def tensor2sam(image):
|
|
# Convert tensor to numpy array in HWC uint8 format with pixel values in [0, 255]
|
|
sam_image = np.clip(255. * image.cpu().numpy().squeeze(), 0, 255).astype(np.uint8)
|
|
# Transpose the image to HWC format if it's in CHW format
|
|
if sam_image.shape[0] == 3:
|
|
sam_image = np.transpose(sam_image, (1, 2, 0))
|
|
return sam_image
|
|
|
|
class SAMLoader:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"model_name": (folder_paths.get_filename_list("sams"), ),
|
|
"device_mode": (["AUTO", "Prefer GPU", "CPU"],),
|
|
}
|
|
}
|
|
|
|
RELOAD_INST = True
|
|
RETURN_TYPES = ("SAM_MODEL",)
|
|
FUNCTION = "load_model"
|
|
|
|
CATEGORY = "Matte Anything"
|
|
|
|
def load_model(self, model_name, device_mode="auto"):
|
|
modelname = folder_paths.get_full_path("sams", model_name)
|
|
|
|
if 'vit_h' in model_name:
|
|
model_kind = 'vit_h'
|
|
elif 'vit_l' in model_name:
|
|
model_kind = 'vit_l'
|
|
else:
|
|
model_kind = 'vit_b'
|
|
|
|
sam = sam_model_registry[model_kind](checkpoint=modelname)
|
|
# Unless user explicitly wants to use CPU, we use GPU
|
|
device = comfy.model_management.get_torch_device() if device_mode == "Prefer GPU" else "CPU"
|
|
|
|
if device_mode == "Prefer GPU":
|
|
sam.to(device=device)
|
|
|
|
sam.is_auto_mode = device_mode == "AUTO"
|
|
print(f"Loads SAM model: {modelname} (device:{device_mode})")
|
|
return (sam,)
|
|
|
|
class InitSAMPredictor:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"image": ("IMAGE", {}),
|
|
"sam_model": ("SAM_MODEL", {}),
|
|
}
|
|
}
|
|
|
|
RELOAD_INST = True
|
|
RETURN_TYPES = ("SAM_PREDICTOR",)
|
|
FUNCTION = "init_predictor"
|
|
|
|
CATEGORY = "Matte Anything"
|
|
|
|
def init_predictor(self, image, sam_model):
|
|
predictor = SamPredictor(sam_model)
|
|
predictor.set_image(tensor2sam(image))
|
|
return (predictor,)
|
|
|
|
# folder_paths.folder_names_and_paths["dinos"] = ([os.path.join(folder_paths.models_dir, "dinos")]
|
|
class LoadDINOModel:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"model_name": (folder_paths.get_filename_list("dino"), ),
|
|
"device_mode": (["AUTO", "Prefer GPU", "CPU"],),
|
|
}
|
|
}
|
|
|
|
RELOAD_INST = True
|
|
RETURN_TYPES = ("DINO_MODEL",)
|
|
FUNCTION = "load_model"
|
|
|
|
CATEGORY = "Matte Anything"
|
|
|
|
def load_model(self, model_name, device_mode="auto"):
|
|
model = DinoModel(
|
|
model_config_path=GroundingDINO_SwinT_OGC.__file__,
|
|
model_checkpoint_path=folder_paths.get_full_path("dino", model_name),
|
|
device="cpu"
|
|
)
|
|
return (model,)
|
|
|
|
def pil2cv(image: Image) -> np.ndarray:
|
|
mode = image.mode
|
|
new_image: np.ndarray
|
|
if mode == '1':
|
|
new_image = np.array(image, dtype=np.uint8)
|
|
new_image *= 255
|
|
elif mode == 'L':
|
|
new_image = np.array(image, dtype=np.uint8)
|
|
elif mode == 'LA' or mode == 'La':
|
|
new_image = np.array(image.convert('RGBA'), dtype=np.uint8)
|
|
new_image = cv2.cvtColor(new_image, cv2.COLOR_RGBA2BGRA)
|
|
elif mode == 'RGB':
|
|
new_image = np.array(image, dtype=np.uint8)
|
|
new_image = cv2.cvtColor(new_image, cv2.COLOR_RGB2BGR)
|
|
elif mode == 'RGBA':
|
|
new_image = np.array(image, dtype=np.uint8)
|
|
new_image = cv2.cvtColor(new_image, cv2.COLOR_RGBA2BGRA)
|
|
elif mode == 'LAB':
|
|
new_image = np.array(image, dtype=np.uint8)
|
|
new_image = cv2.cvtColor(new_image, cv2.COLOR_LAB2BGR)
|
|
elif mode == 'HSV':
|
|
new_image = np.array(image, dtype=np.uint8)
|
|
new_image = cv2.cvtColor(new_image, cv2.COLOR_HSV2BGR)
|
|
elif mode == 'YCbCr':
|
|
# XXX: not sure if YCbCr == YCrCb
|
|
new_image = np.array(image, dtype=np.uint8)
|
|
new_image = cv2.cvtColor(new_image, cv2.COLOR_YCrCb2BGR)
|
|
elif mode == 'P' or mode == 'CMYK':
|
|
new_image = np.array(image.convert('RGB'), dtype=np.uint8)
|
|
new_image = cv2.cvtColor(new_image, cv2.COLOR_RGB2BGR)
|
|
elif mode == 'PA' or mode == 'Pa':
|
|
new_image = np.array(image.convert('RGBA'), dtype=np.uint8)
|
|
new_image = cv2.cvtColor(new_image, cv2.COLOR_RGBA2BGRA)
|
|
else:
|
|
raise ValueError(f'unhandled image color mode: {mode}')
|
|
|
|
return new_image
|
|
|
|
def tensor2pil(image: Tensor) -> PIL.Image.Image:
|
|
return Image.fromarray(np.clip(255. * image.cpu().numpy().squeeze(), 0, 255).astype(np.uint8))
|
|
|
|
def pil2tensor(image: PIL.Image.Image) -> Tensor:
|
|
return torch.from_numpy(np.array(image).astype(np.float32) / 255.0).unsqueeze(0)
|
|
class DinoBoxes:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"image": ("IMAGE", {}),
|
|
"dino_model": ("DINO_MODEL", {}),
|
|
"caption": ("STRING", {}),
|
|
}
|
|
}
|
|
|
|
RELOAD_INST = True
|
|
RETURN_TYPES = ("IMAGE", "DINO_BOXES")
|
|
FUNCTION = "get_boxes"
|
|
|
|
CATEGORY = "Matte Anything"
|
|
|
|
def get_boxes(self, image, dino_model: DinoModel, caption: str):
|
|
point_coords, point_labels = None, None
|
|
image_cv = pil2cv((tensor2pil(image)))
|
|
detections, labels = dino_model.predict_with_caption(
|
|
image=image_cv,
|
|
caption=caption,
|
|
)
|
|
# Detection is
|
|
"""
|
|
Data class containing information about the detections in a video frame.
|
|
Attributes:
|
|
xyxy (np.ndarray): An array of shape `(n, 4)` containing the bounding boxes coordinates in format `[x1, y1, x2, y2]`
|
|
mask: (Optional[np.ndarray]): An array of shape `(n, W, H)` containing the segmentation masks.
|
|
confidence (Optional[np.ndarray]): An array of shape `(n,)` containing the confidence scores of the detections.
|
|
class_id (Optional[np.ndarray]): An array of shape `(n,)` containing the class ids of the detections.
|
|
tracker_id (Optional[np.ndarray]): An array of shape `(n,)` containing the tracker ids of the detections.
|
|
"""
|
|
image_pil = tensor2pil(image)
|
|
img_draw = ImageDraw.Draw(image_pil)
|
|
for detection in detections:
|
|
xyxy = detection[0]
|
|
img_draw.rectangle(xyxy, outline="red", width=4)
|
|
return (pil2tensor(image_pil),detections)
|
|
|
|
class SAMMaskFromBoxes:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"sam_predictor": ("SAM_PREDICTOR", {}),
|
|
"boxes": ("DINO_BOXES", {}),
|
|
}
|
|
}
|
|
|
|
RELOAD_INST = True
|
|
RETURN_TYPES = ("MASK",)
|
|
FUNCTION = "get_mask"
|
|
|
|
CATEGORY = "Matte Anything"
|
|
|
|
def get_mask(self, sam_predictor: SamPredictor, boxes):
|
|
boxes_stacked = torch.stack([torch.tensor(box[0]) for box in boxes])
|
|
masks, scores, logits = sam_predictor.predict_torch(
|
|
point_coords=None,
|
|
point_labels=None,
|
|
boxes=boxes_stacked,
|
|
multimask_output=False
|
|
)
|
|
|
|
return (torch.squeeze(masks),)
|
|
|
|
def generate_trimap(mask, erode_kernel_size=10, dilate_kernel_size=10):
|
|
erode_kernel = np.ones((erode_kernel_size, erode_kernel_size), np.uint8)
|
|
dilate_kernel = np.ones((dilate_kernel_size, dilate_kernel_size), np.uint8)
|
|
eroded = cv2.erode(mask, erode_kernel, iterations=5)
|
|
dilated = cv2.dilate(mask, dilate_kernel, iterations=5)
|
|
trimap = np.zeros_like(mask)
|
|
trimap[dilated==255] = 128
|
|
trimap[eroded==255] = 255
|
|
return trimap
|
|
|
|
class MaskToTrimap:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"mask": ("MASK", {}),
|
|
"erode_kernel_size": ("INT", {"min": 1, "step": 1}),
|
|
"dilate_kernel_size": ("INT", {"min": 1, "step": 1}),
|
|
}
|
|
}
|
|
|
|
RELOAD_INST = True
|
|
RETURN_TYPES = ("TRIMAP","MASK")
|
|
FUNCTION = "get_trimap"
|
|
|
|
CATEGORY = "Matte Anything"
|
|
|
|
def get_trimap(self, mask: Tensor, erode_kernel_size: int, dilate_kernel_size: int):
|
|
mask = mask.cpu().detach().numpy().astype(np.uint8)*255
|
|
trimap = generate_trimap(mask, erode_kernel_size, dilate_kernel_size).astype(np.float32)
|
|
trimap[trimap==128] = 0.5
|
|
trimap[trimap==255] = 1
|
|
trimap = torch.from_numpy(trimap).unsqueeze(0)
|
|
# mask_np = mask.cpu().numpy()
|
|
# mask_np = np.clip(mask_np, 0, 1)
|
|
# mask_np = mask_np.astype(np.uint8)
|
|
# mask_np = mask_np * 255
|
|
# mask_np = cv2.GaussianBlur(mask_np, (5, 5), 0)
|
|
# mask_np = mask_np.astype(np.float32) / 255.0
|
|
# trimap = torch.from_numpy(mask_np).unsqueeze(0)
|
|
return (trimap,trimap)
|
|
|
|
|
|
vitmatte_config = {
|
|
'vit_b': './custom_nodes/Comfy_MatteAnything/Matte_Anything/configs/matte_anything.py',
|
|
}
|
|
def init_vitmatte(model_type, model_path):
|
|
"""
|
|
Initialize the vitmatte with model_type in ['vit_s', 'vit_b']
|
|
"""
|
|
cfg = LazyConfig.load(os.path.abspath(vitmatte_config[model_type]))
|
|
vitmatte = instantiate(cfg.model)
|
|
device = comfy.model_management.get_torch_device()
|
|
vitmatte.to(device)
|
|
vitmatte.eval()
|
|
DetectionCheckpointer(vitmatte).load(model_path)
|
|
|
|
return vitmatte
|
|
|
|
|
|
class LoadVITMatteModel:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"model_name": (folder_paths.get_filename_list("matte"), ),
|
|
"device_mode": (["AUTO", "Prefer GPU", "CPU"],),
|
|
}
|
|
}
|
|
|
|
RELOAD_INST = True
|
|
RETURN_TYPES = ("VIT_MATTE_MODEL",)
|
|
FUNCTION = "load_model"
|
|
|
|
CATEGORY = "Matte Anything"
|
|
|
|
def load_model(self, model_name, device_mode="auto"):
|
|
model_path = folder_paths.get_full_path("matte", model_name)
|
|
vitmatte = init_vitmatte('vit_b', model_path)
|
|
return (vitmatte,)
|
|
|
|
class GenerateVITMatte:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"image": ("IMAGE", {}),
|
|
"trimap": ("TRIMAP", {}),
|
|
"vit_matte_model": ("VIT_MATTE_MODEL", {}),
|
|
}
|
|
}
|
|
|
|
RELOAD_INST = True
|
|
RETURN_TYPES = ("IMAGE",)
|
|
FUNCTION = "generate_matte"
|
|
|
|
CATEGORY = "Matte Anything"
|
|
|
|
def generate_matte(self, image, trimap, vit_matte_model):
|
|
image_in = pil2cv(tensor2pil(image))
|
|
device = comfy.model_management.get_torch_device()
|
|
input = {
|
|
"image": torch.from_numpy(image_in).permute(2, 0, 1).unsqueeze(0).to(device)/255,
|
|
"trimap": trimap.unsqueeze(0).to(device),
|
|
# "trimap": torch.from_numpy(trimap).unsqueeze(0).unsqueeze(0),
|
|
}
|
|
alpha = vit_matte_model(input)['phas'].flatten(0, 2)
|
|
alpha = alpha.detach().cpu().numpy()
|
|
# Converts alpha matte to RGBA image using the image tensor
|
|
image = cv2.cvtColor(image_in, cv2.COLOR_BGR2RGBA)
|
|
alpha = np.clip(255. * alpha, 0, 255).astype(np.uint8)
|
|
image[:, :, 3] = alpha
|
|
image = Image.fromarray(image)
|
|
image = pil2tensor(image)
|
|
return (image,)
|