88 lines
3.2 KiB
Python
88 lines
3.2 KiB
Python
import os
|
|
import torch
|
|
from PIL import Image
|
|
import torchvision.transforms.functional as tf
|
|
from .model import Harmonizer # Harmonizer 모델 import (경로에 따라 수정 필요)
|
|
import logging
|
|
|
|
class HarmonizerNode:
|
|
def __init__(self):
|
|
self.harmonizer = None
|
|
self.cuda = torch.cuda.is_available()
|
|
self.load_model()
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"composite_image": ("IMAGE",),
|
|
"mask_image": ("IMAGE",),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE",)
|
|
RETURN_NAMES = ("harmonized_image",)
|
|
FUNCTION = "harmonize_image"
|
|
CATEGORY = "image"
|
|
|
|
def load_model(self):
|
|
try:
|
|
self.harmonizer = Harmonizer()
|
|
if self.cuda:
|
|
self.harmonizer = self.harmonizer.cuda()
|
|
model_path = os.path.join(os.path.dirname(__file__), 'harmonizer.pth')
|
|
self.harmonizer.load_state_dict(torch.load(model_path), strict=True)
|
|
self.harmonizer.eval()
|
|
except Exception as e:
|
|
logging.exception("Failed to load Harmonizer model.")
|
|
raise e
|
|
|
|
def harmonize_image(self, composite_image, mask_image, *args, **kwargs):
|
|
try:
|
|
# Convert composite_image to (N, C, H, W)
|
|
composite_image = composite_image.permute(0, 3, 1, 2) # From (N, H, W, C) to (N, C, H, W)
|
|
if composite_image.shape[1] != 3:
|
|
raise ValueError(f"Expected composite_image to have 3 channels (RGB), but got {composite_image.shape[1]} channels.")
|
|
|
|
# Convert mask_image to (N, C, H, W) and ensure single channel
|
|
mask_image = mask_image.permute(0, 3, 1, 2) # From (N, H, W, C) to (N, C, H, W)
|
|
if mask_image.shape[1] != 1:
|
|
mask_image = mask_image[:, :1, :, :] # Keep only the first channel
|
|
|
|
# Move to CUDA if available
|
|
if self.cuda:
|
|
composite_image, mask_image = composite_image.cuda(), mask_image.cuda()
|
|
|
|
# Perform harmonization
|
|
with torch.no_grad():
|
|
arguments = self.harmonizer.predict_arguments(composite_image, mask_image)
|
|
harmonized = self.harmonizer.restore_image(composite_image, mask_image, arguments)[-1]
|
|
|
|
# Convert harmonized tensor back to PIL Image
|
|
harmonized_image = (harmonized.squeeze(0).permute(1, 2, 0).cpu().numpy() * 255).clip(0, 255).astype('uint8')
|
|
harmonized_pil = Image.fromarray(harmonized_image)
|
|
|
|
# Use PIL.Image size for logging or other operations
|
|
image_size = harmonized_pil.size # (Width, Height)
|
|
harmonized_pil = tf.to_tensor(harmonized_pil)
|
|
harmonized_pil = harmonized_pil.unsqueeze(dim=0)
|
|
harmonized_pil = harmonized_pil.permute(0, 2, 3, 1) # From (N, H, W, C) to (N, C, H, W)
|
|
|
|
|
|
# Return the harmonized image as output
|
|
return (harmonized_pil,)
|
|
except Exception as e:
|
|
logging.exception("Error during harmonization.")
|
|
raise e
|
|
|
|
|
|
|
|
|
|
NODE_CLASS_MAPPINGS = {
|
|
"harmonizer": HarmonizerNode,
|
|
}
|
|
|
|
NODE_DISPLAY_NAME_MAPPINGS = {
|
|
"harmonizer": "Image Harmonizer",
|
|
}
|