Update main_code_script.py

main_code_script.py

@@ -5,99 +5,196 @@ from PIL import Image
 import cv2
 import mediapipe as mp
 import numpy as np
-from transformers import pipeline
-from diffusers import StableDiffusionInpaintPipeline
+from transformers import pipeline, CLIPImageProcessor, CLIPVisionModelWithProjection, CLIPTextModel, CLIPTextModelWithProjection, AutoTokenizer
+from diffusers import StableDiffusionXLInpaintPipeline, DDPMScheduler, AutoencoderKL
 import torch
-# --- 1. Pose Estimation (using Mediapipe) ---
-def estimate_pose(image_path):
-    """Detects the pose of a person in an image using Mediapipe.
-    Args:
-        image_path: Path to the input image.
-    Returns:
-        A list of landmarks (x, y, visibility) 
-        or None if no pose is detected.
-    """
-    mp_drawing = mp.solutions.drawing_utils
-    mp_pose = mp.solutions.pose
-    with mp_pose.Pose(
-        static_image_mode=True,
-        model_complexity=2,
-        enable_segmentation=True,
-        min_detection_confidence=0.5) as pose:
-        image = cv2.imread(image_path)
-        image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
-        results = pose.process(image_rgb)
-        
-    if results.pose_landmarks:
-        # Example: Draw the pose landmarks on the image (for visualization)
-        annotated_image = image.copy()
-        mp_drawing.draw_landmarks(
-            annotated_image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)
-        #cv2.imwrite("pose_annotated.jpg", annotated_image)  # Save annotated image
-        #return results.pose_landmarks.landmark  
-        # Return the landmarks
-        return results, image # Return the entire result
+import os
+from torchvision import transforms
+from typing import List
+# from utils_mask import get_mask_location
+# from preprocess.humanparsing.run_parsing import Parsing
+# from preprocess.openpose.run_openpose import OpenPose
+from detectron2.data.detection_utils import convert_PIL_to_numpy, _apply_exif_orientation
+from torchvision.transforms.functional import to_pil_image
+import apply_net
+
+def pil_to_binary_mask(pil_image, threshold=0):
+    np_image = np.array(pil_image)
+    grayscale_image = Image.fromarray(np_image).convert("L")
+    binary_mask = np.array(grayscale_image) > threshold
+    mask = np.zeros(binary_mask.shape, dtype=np.uint8)
+    for i in range(binary_mask.shape[0]):
+        for j in range(binary_mask.shape[1]):
+            if binary_mask[i, j] == True:
+                mask[i, j] = 1
+    mask = (mask * 255).astype(np.uint8)
+    output_mask = Image.fromarray(mask)
+    return output_mask # [cite: 60, 61]
+base_path = 'yisol/IDM-VTON'
+# example_path = os.path.join(os.path.dirname(__file__), 'example')
+unet = UNet2DConditionModel.from_pretrained(
+    base_path,
+    subfolder="unet",
+    torch_dtype=torch.float16,
+)
+unet.requires_grad_(False) # [cite: 61]
+tokenizer_one = AutoTokenizer.from_pretrained(
+    base_path,
+    subfolder="tokenizer",
+    revision=None,
+    use_fast=False,
+) # [cite: 61, 62]
+tokenizer_two = AutoTokenizer.from_pretrained(
+    base_path,
+    subfolder="tokenizer_2",
+    revision=None,
+    use_fast=False,
+) # [cite: 62]
+noise_scheduler = DDPMScheduler.from_pretrained(base_path, subfolder="scheduler") # [cite: 62]
+text_encoder_one = CLIPTextModel.from_pretrained(
+    base_path,
+    subfolder="text_encoder",
+    torch_dtype=torch.float16,
+) # [cite: 62]
+text_encoder_two = CLIPTextModelWithProjection.from_pretrained(
+    base_path,
+    subfolder="text_encoder_2",
+    torch_dtype=torch.float16,
+) # [cite: 62]
+image_encoder = CLIPVisionModelWithProjection.from_pretrained(
+    base_path,
+    subfolder="image_encoder",
+    torch_dtype=torch.float16,
+) # [cite: 62, 63]
+vae = AutoencoderKL.from_pretrained(base_path,
+                                    subfolder="vae",
+                                    torch_dtype=torch.float16,
+) # [cite: 63]
+# "stabilityai/stable-diffusion-xl-base-1.0",
+UNet_Encoder = UNet2DConditionModel_ref.from_pretrained(
+    base_path,
+    subfolder="unet_encoder",
+    torch_dtype=torch.float16,
+) # [cite: 63]
+# parsing_model = Parsing(0)
+# openpose_model = OpenPose(0)
+UNet_Encoder.requires_grad_(False) # [cite: 63]
+image_encoder.requires_grad_(False) # [cite: 63]
+vae.requires_grad_(False) # [cite: 63]
+unet.requires_grad_(False) # [cite: 63]
+text_encoder_one.requires_grad_(False) # [cite: 63]
+text_encoder_two.requires_grad_(False) # [cite: 63]
+tensor_transfrom = transforms.Compose(
+    [
+        transforms.ToTensor(),
+        transforms.Normalize(,),
+    ]
+) # [cite: 63, 64]
+pipe = TryonPipeline.from_pretrained(
+    base_path,
+    unet=unet,
+    vae=vae,
+    feature_extractor=CLIPImageProcessor(),
+    text_encoder=text_encoder_one,
+    text_encoder_2=text_encoder_two,
+    tokenizer=tokenizer_one,
+    tokenizer_2=tokenizer_two,
+    scheduler=noise_scheduler,
+    image_encoder=image_encoder,
+    torch_dtype=torch.float16,
+) # [cite: 64, 65]
+pipe.unet_encoder = UNet_Encoder # [cite: 65]
+@spaces.GPU
+def start_tryon(dict, garm_img, garment_des, is_checked, is_checked_crop, denoise_steps, seed):
+    device = "cuda"
+    # openpose_model.preprocessor.body_estimation.model.to(device)
+    pipe.to(device)
+    pipe.unet_encoder.to(device)
+    garm_img = garm_img.convert("RGB").resize((768, 1024))
+    human_img_orig = dict["background"].convert("RGB")
+    if is_checked_crop:
+        width, height = human_img_orig.size
+        target_width = int(min(width, height * (3 / 4)))
+        target_height = int(min(height, width * (4 / 3)))
+        left = (width - target_width) / 2
+        top = (height - target_height) / 2
+        right = (width + target_width) / 2
+        bottom = (height + target_height) / 2
+        cropped_img = human_img_orig.crop((left, top, right, bottom))
+        crop_size = cropped_img.size
+        human_img = cropped_img.resize((384, 512))  # Reduced size for efficiency
     else:
-        return None, None # or raise an exception
-# --- 2. Clothing Segmentation (Example - using a placeholder function) ---
-def segment_clothing(image, results): #Added result
-    """Segments the clothing region in the image.
-    This is a simplified example. In reality, you would use a pre-trained
-       segmentation model.
-    """
-    # 1. Create a mask where the person is present.
-    segmentation_mask = results.segmentation_mask
-    threshold = 0.5  # Adjust this threshold as needed.
-    # Threshold the segmentation mask to create a binary mask.
-    binary_mask = (segmentation_mask > threshold).astype(np.uint8) * 255
-    # Convert binary mask to a PIL Image
-    mask_img = Image.fromarray(binary_mask).convert("L")
-    return mask_img
-# --- 3. Image Inpainting (Replacing Clothing - using Stable Diffusion Inpainting) ---
-def inpaint_clothing(image, mask_img, clothing_prompt, device="cuda" if torch.cuda.is_available() else "cpu"):
-    """
-    Replaces the clothing region in the image with new clothing based on a text prompt,
-    using Stable Diffusion Inpainting.
-    """
-    pipe = StableDiffusionInpaintPipeline.from_pretrained(
-        "stabilityai/stable-diffusion-2-inpainting",
-        torch_dtype=torch.float16
-    )
-    pipe = pipe.to(device)
-    # Resize the image and mask to a smaller size for faster inpainting
-    width, height = image.size
-    inpainted_size = (256, 256)  # Smaller size for faster inpainting
-    image = image.resize(inpainted_size)
-    mask_img = mask_img.resize(inpainted_size)
-    prompt = f"A photo of a person wearing {clothing_prompt}" #Add style or detail
-    image = pipe(prompt=prompt, image=image, mask_image=mask_img).images[0]
-    # Resize back to the original size (or a desired output size)
-    image = image.resize((width, height)) # Or resize to a target output size
-    return image
-# --- 4. Main Function (Putting it all together) ---
-def change_clothing(image_path, clothing_prompt):
-    """
-    Main function to change the clothing in an image.
-    """
-    # 1. Load the image
-    image = Image.open(image_path).convert("RGB")
-    # 2. Estimate the pose
-    results, cv2_image = estimate_pose(image_path)
-    if results is None:
-        print("No pose detected.")
-        return None
-    # 3. Segment the clothing
-    mask_img = segment_clothing(image, results)
-    # 4. Inpaint the clothing
-    modified_image = inpaint_clothing(image, mask_img, clothing_prompt)
-    return modified_image
-# --- Example Usage ---
-if __name__ == "__main__":
-    input_image_path = "person.jpg"  # Replace with your image
-    clothing_description = "a red leather jacket"  # Replace with desired clothing
-    modified_image = change_clothing(input_image_path, clothing_description)
-    if modified_image:
-        modified_image.save("modified_image.jpg")
-        print("Clothing changed and saved to modified_image.jpg")
+        human_img = human_img_orig.resize((384, 512))  # Reduced size for efficiency
+    if is_checked:
+        # keypoints = openpose_model(human_img.resize((384, 512)))
+        # model_parse, _ = parsing_model(human_img.resize((384, 512)))
+        # mask, mask_gray = get_mask_location('hd', "upper_body", model_parse, keypoints)
+        # mask = mask.resize((768, 1024))
+        # Placeholder for mask generation (replace with your mask logic)
+        mask = Image.new('L', (768, 1024), color='white')  # Example: a white mask
+        mask_gray = Image.new('RGB', (768, 1024), color='gray') # Example: a gray image
     else:
-        print("Failed to change clothing.")
+        mask = pil_to_binary_mask(dict['layers'][0].convert("RGB").resize((768, 1024)))
+    mask_gray = (1 - transforms.ToTensor()(mask)) * tensor_transfrom(human_img)
+    mask_gray = to_pil_image((mask_gray + 1.0) / 2.0)
+    human_img_arg = _apply_exif_orientation(human_img.resize((384, 512)))
+    human_img_arg = convert_PIL_to_numpy(human_img_arg, format="BGR")
+    args = apply_net.create_argument_parser().parse_args(
+        ('show', './configs/densepose_rcnn_R_50_FPN_s1x.yaml', './ckpt/densepose/model_final_162be9.pkl', 'dp_segm',
+         '-v', '--opts', 'MODEL.DEVICE', 'cuda'))
+    # verbosity = getattr(args, "verbosity", None)
+    pose_img = args.func(args, human_img_arg)
+    pose_img = pose_img[:, :, ::-1]
+    pose_img = Image.fromarray(pose_img).resize((768, 1024))
+    with torch.no_grad():
+        # Extract the images
+        with torch.cuda.amp.autocast():
+            with torch.no_grad():
+                prompt = "model is wearing " + garment_des
+                negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+                with torch.inference_mode():
+                    (
+                        prompt_embeds,
+                        negative_prompt_embeds,
+                        pooled_prompt_embeds,
+                        negative_pooled_prompt_embeds,
+                    ) = pipe.encode_prompt(
+                        prompt,
+                        num_images_per_prompt=1,
+                        do_classifier_free_guidance=True,
+                        negative_prompt=negative_prompt,
+                    )
+                prompt = "a photo of " + garment_des
+                negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+                if not isinstance(prompt, List):
+                    prompt = [prompt] * 1
+                if not isinstance(negative_prompt, List):
+                    negative_prompt = [negative_prompt] * 1
+                with torch.inference_mode():
+                    (
+                        prompt_embeds_c,
+                        _,
+                        _,
+                        _,
+                    ) = pipe.encode_prompt(
+                        prompt,
+                        num_images_per_prompt=1,
+                        do_classifier_free_guidance=False,
+                        negative_prompt=negative_prompt,
+                    )
+                pose_img = tensor_transfrom(pose_img).unsqueeze(0).to(device, torch.float16)
+                garm_tensor = tensor_transfrom(garm_img).unsqueeze(0).to(device, torch.float16)
+                generator = torch.Generator(device).manual_seed(seed) if seed is not None else None
+                images = pipe(
+                    prompt_embeds=prompt_embeds.to(device, torch.float16),
+                    negative_prompt_embeds=negative_prompt_embeds.to(device, torch.float16),
+                    pooled_prompt_embeds=pooled_prompt_embeds.to(device, torch.float16),
+                    negative_pooled_prompt_embeds=negative_pooled_prompt_embeds.to(device, torch.float16),
+                    num_inference_steps=denoise_steps, # [cite: 78, 79, 80]
+                    generator=generator,
+                    strength=1.0,
+                    pose_img=pose_img.to(device, torch.float16),
+                    text_embeds_cloth=prompt_embeds_c.to(device, torch.float16),
+                    cloth=garm_tensor.to(device, torch.float16),
+                    mask_image=mask,
+                    image=human_