Update app.py

app.py

@@ -1,21 +1,27 @@
 import os
-from flask import Flask, request, jsonify, send_file
+from flask import Flask, request, jsonify,send_file
 from PIL import Image
 from io import BytesIO
-import base64
 import torch
-import requests
+import base64
+import io 
+import logging
+import gradio as gr
 import numpy as np
-import uuid
 import spaces
+import uuid
+import random
+from src.tryon_pipeline import StableDiffusionXLInpaintPipeline as TryonPipeline
+from src.unet_hacked_garmnet import UNet2DConditionModel as UNet2DConditionModel_ref
+from src.unet_hacked_tryon import UNet2DConditionModel
 from transformers import (
     CLIPImageProcessor,
     CLIPVisionModelWithProjection,
     CLIPTextModel,
     CLIPTextModelWithProjection,
-    AutoTokenizer
+    AutoTokenizer,
 )
-from diffusers import DDPMScheduler, AutoencoderKL, UNet2DConditionModel
+from diffusers import DDPMScheduler, AutoencoderKL
 from utils_mask import get_mask_location
 from torchvision import transforms
 import apply_net
@@ -26,78 +32,114 @@ from torchvision.transforms.functional import to_pil_image
 
 app = Flask(__name__)
 
-# Variables globales pour stocker les modèles
-models_loaded = False
+base_path = 'yisol/IDM-VTON'
+example_path = os.path.join(os.path.dirname(__file__), 'example')
 
-def load_models():
-    global unet, tokenizer_one, tokenizer_two, noise_scheduler, text_encoder_one, text_encoder_two
-    global image_encoder, vae, UNet_Encoder, parsing_model, openpose_model, pipe
-    global models_loaded
-    
-    if not models_loaded:
-        base_path = 'yisol/IDM-VTON'
-        unet = UNet2DConditionModel.from_pretrained(base_path, subfolder="unet", torch_dtype=torch.float16, force_download=False)
-        unet.requires_grad_(False)
-        
-        tokenizer_one = AutoTokenizer.from_pretrained(base_path, subfolder="tokenizer", use_fast=False, force_download=False)
-        tokenizer_two = AutoTokenizer.from_pretrained(base_path, subfolder="tokenizer_2", use_fast=False, force_download=False)
-        
-        noise_scheduler = DDPMScheduler.from_pretrained(base_path, subfolder="scheduler")
-        text_encoder_one = CLIPTextModel.from_pretrained(base_path, subfolder="text_encoder", torch_dtype=torch.float16, force_download=False)
-        text_encoder_two = CLIPTextModelWithProjection.from_pretrained(base_path, subfolder="text_encoder_2", torch_dtype=torch.float16, force_download=False)
-        image_encoder = CLIPVisionModelWithProjection.from_pretrained(base_path, subfolder="image_encoder", torch_dtype=torch.float16, force_download=False)
-        vae = AutoencoderKL.from_pretrained(base_path, subfolder="vae", torch_dtype=torch.float16, force_download=False)
-        
-        # Set the correct encoder_hid_dim_type here
-        UNet_Encoder = UNet2DConditionModel.from_pretrained(
-            base_path,
-            subfolder="unet_encoder",
-            torch_dtype=torch.float16,
-            encoder_hid_dim_type="text_proj",  # Update based on model type
-            force_download=False
-        )
-        
-        parsing_model = Parsing(0)
-        openpose_model = OpenPose(0)
-        
-        UNet_Encoder.requires_grad_(False)
-        image_encoder.requires_grad_(False)
-        vae.requires_grad_(False)
-        unet.requires_grad_(False)
-        text_encoder_one.requires_grad_(False)
-        text_encoder_two.requires_grad_(False)
-        
-        tensor_transfrom = transforms.Compose([transforms.ToTensor(), transforms.Normalize([0.5], [0.5])])
-        
-        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,
-            force_download=False
-        )
-        pipe.unet_encoder = UNet_Encoder
-        
-        models_loaded = True
+unet = UNet2DConditionModel.from_pretrained(
+    base_path,
+    subfolder="unet",
+    torch_dtype=torch.float16,
+    force_download=False
+)
+unet.requires_grad_(False)
+tokenizer_one = AutoTokenizer.from_pretrained(
+    base_path,
+    subfolder="tokenizer",
+    revision=None,
+    use_fast=False,
+    force_download=False
+)
+tokenizer_two = AutoTokenizer.from_pretrained(
+    base_path,
+    subfolder="tokenizer_2",
+    revision=None,
+    use_fast=False,
+    force_download=False
+)
+noise_scheduler = DDPMScheduler.from_pretrained(base_path, subfolder="scheduler")
+
+text_encoder_one = CLIPTextModel.from_pretrained(
+    base_path,
+    subfolder="text_encoder",
+    torch_dtype=torch.float16,
+    force_download=False
+)
+text_encoder_two = CLIPTextModelWithProjection.from_pretrained(
+    base_path,
+    subfolder="text_encoder_2",
+    torch_dtype=torch.float16,
+    force_download=False
+)
+image_encoder = CLIPVisionModelWithProjection.from_pretrained(
+    base_path,
+    subfolder="image_encoder",
+    torch_dtype=torch.float16,
+    force_download=False
+)
+vae = AutoencoderKL.from_pretrained(base_path,
+                                    subfolder="vae",
+                                    torch_dtype=torch.float16,
+                                    force_download=False
+)
+
+UNet_Encoder = UNet2DConditionModel_ref.from_pretrained(
+    base_path,
+    subfolder="unet_encoder",
+    torch_dtype=torch.float16,
+    force_download=False
+)
+
+parsing_model = Parsing(0)
+openpose_model = OpenPose(0)
+
+UNet_Encoder.requires_grad_(False)
+image_encoder.requires_grad_(False)
+vae.requires_grad_(False)
+unet.requires_grad_(False)
+text_encoder_one.requires_grad_(False)
+text_encoder_two.requires_grad_(False)
+tensor_transfrom = transforms.Compose(
+            [
+                transforms.ToTensor(),
+                transforms.Normalize([0.5], [0.5]),
+            ]
+    )
+
+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,
+        force_download=False
+)
+pipe.unet_encoder = UNet_Encoder
 
 def pil_to_binary_mask(pil_image, threshold=0):
-    np_image = np.array(pil_image.convert("L"))  # Convert to grayscale directly
-    binary_mask = np_image > threshold
-    mask = np.uint8(binary_mask * 255)
-    return Image.fromarray(mask)
+    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]:
+                mask[i, j] = 1
+    mask = (mask * 255).astype(np.uint8)
+    output_mask = Image.fromarray(mask)
+    return output_mask
 
 def get_image_from_url(url):
     try:
         response = requests.get(url)
-        response.raise_for_status()
-        return Image.open(BytesIO(response.content))
+        response.raise_for_status()  # Vérifie les erreurs HTTP
+        img = Image.open(BytesIO(response.content))
+        return img
     except Exception as e:
         logging.error(f"Error fetching image from URL: {e}")
         raise
@@ -105,7 +147,8 @@ def get_image_from_url(url):
 def decode_image_from_base64(base64_str):
     try:
         img_data = base64.b64decode(base64_str)
-        return Image.open(BytesIO(img_data))
+        img = Image.open(BytesIO(img_data))
+        return img
     except Exception as e:
         logging.error(f"Error decoding image: {e}")
         raise
@@ -114,142 +157,257 @@ def encode_image_to_base64(img):
     try:
         buffered = BytesIO()
         img.save(buffered, format="PNG")
-        return base64.b64encode(buffered.getvalue()).decode("utf-8")
+        img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
+        return img_str
     except Exception as e:
         logging.error(f"Error encoding image: {e}")
         raise
 
 def save_image(img):
-    unique_name = f"{uuid.uuid4()}.webp"
-    img.save(unique_name, format="WEBP", lossless=True)
+    unique_name = str(uuid.uuid4()) + ".webp"  
+    img.save(unique_name, format="WEBP", lossless=True) 
     return unique_name
 
-def clear_gpu_memory():
-    torch.cuda.empty_cache()
-    torch.cuda.ipc_collect()
-
 @spaces.GPU
-def start_tryon(human_dict, garment_image, garment_description, use_auto_mask, use_auto_crop, denoise_steps, seed, category='upper_body'):
+def start_tryon(dict, garm_img, garment_des, is_checked, is_checked_crop, denoise_steps, seed, categorie = 'upper_body'):
     device = "cuda"
     openpose_model.preprocessor.body_estimation.model.to(device)
     pipe.to(device)
     pipe.unet_encoder.to(device)
 
-    garment_image = garment_image.convert("RGB").resize((768, 1024))
-    human_image_orig = human_dict["background"].convert("RGB")
+    garm_img = garm_img.convert("RGB").resize((768, 1024))
+    human_img_orig = dict["background"].convert("RGB")
 
-    if use_auto_crop:
-        width, height = human_image_orig.size
+    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, top = (width - target_width) / 2, (height - target_height) / 2
-        right, bottom = (width + target_width) / 2, (height + target_height) / 2
-        cropped_img = human_image_orig.crop((left, top, right, bottom)).resize((768, 1024))
+        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((768, 1024))
     else:
-        cropped_img = human_image_orig.resize((768, 1024))
+        human_img = human_img_orig.resize((768, 1024))
 
-    if use_auto_mask:
-        keypoints = openpose_model(cropped_img.resize((384, 512)))
-        model_parse, _ = parsing_model(cropped_img.resize((384, 512)))
-        mask, mask_gray = get_mask_location('hd', category, model_parse, keypoints)
+    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', categorie , model_parse, keypoints)
         mask = mask.resize((768, 1024))
     else:
-        mask = pil_to_binary_mask(human_dict['layers'][0].convert("RGB").resize((768, 1024)))
-    
-    mask_gray = (1 - transforms.ToTensor()(mask)) * transforms.Compose([transforms.ToTensor(), transforms.Normalize([0.5], [0.5])])(cropped_img)
+        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_image_arg = _apply_exif_orientation(cropped_img.resize((384, 512)))
-    human_image_arg = convert_PIL_to_numpy(human_image_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'))
-    pose_image = args.func(args, human_image_arg)
-    pose_image = Image.fromarray(pose_image[:, :, ::-1]).resize((768, 1024))
-
-    with torch.no_grad(), torch.cuda.amp.autocast():
-        prompt = "model is wearing " + garment_description
-        negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
-        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_c = "a photo of " + garment_description
-        negative_prompt_c = "monochrome, lowres, bad anatomy, worst quality, low quality"
-        prompt_embeds_c, _, _, _ = pipe.encode_prompt(
-            prompt_c, num_images_per_prompt=1, do_classifier_free_guidance=False, negative_prompt=negative_prompt_c
-        )
-
-        pose_image = transforms.Compose([transforms.ToTensor(), transforms.Normalize([0.5], [0.5])])(pose_image).unsqueeze(0).to(device, torch.float16)
-        garment_tensor = transforms.Compose([transforms.ToTensor(), transforms.Normalize([0.5], [0.5])])(garment_image).unsqueeze(0).to(device, torch.float16)
-
-        images = pipe(
-            prompt_embeds=prompt_embeds.to(device, torch.float16),
-            negative_prompt_embeds=negative_prompt_embeds.to(device, torch.float16),
-            pose_image=pose_image,
-            garment_image=garment_tensor,
-            mask_image=mask_gray.to(device, torch.float16),
-            generator=torch.Generator(device).manual_seed(seed),
-            num_inference_steps=denoise_steps
-        ).images
-
-    if images:
-        output_image = images[0]
-        output_base64 = encode_image_to_base64(output_image)
-        mask_image = mask
-        mask_base64 = encode_image_to_base64(mask_image)
-        return output_image, mask_image
+    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'))
+    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():
+        with torch.cuda.amp.autocast():
+            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,
+                    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_img,
+                    height=1024,
+                    width=768,
+                    ip_adapter_image=garm_img.resize((768, 1024)),
+                    guidance_scale=2.0,
+                )[0]
+
+    if is_checked_crop:
+        out_img = images[0].resize(crop_size)
+        human_img_orig.paste(out_img, (int(left), int(top)))
+        return human_img_orig, mask_gray
     else:
-        raise ValueError("Failed to generate image")
+        return images[0], mask_gray
 
 
-# Route pour récupérer l'image générée
-@app.route('/api/get_image/<image_id>', methods=['GET'])
-def get_image(image_id):
-    # Construire le chemin complet de l'image
-    image_path = image_id  # Assurez-vous que le nom de fichier correspond à celui que vous avez utilisé lors de la sauvegarde
+def clear_gpu_memory():
+    torch.cuda.empty_cache()
+    torch.cuda.synchronize()
 
-    # Renvoyer l'image
-    try:
-        return send_file(image_path, mimetype='image/webp')
-    except FileNotFoundError:
-        return jsonify({'error': 'Image not found'}), 404
+def process_image(image_data):
+    # Vérifie si l'image est en base64 ou URL
+    if image_data.startswith('http://') or image_data.startswith('https://'):
+        return get_image_from_url(image_data)  # Télécharge l'image depuis l'URL
+    else:
+        return decode_image_from_base64(image_data)  # Décode l'image base64
 
 @app.route('/tryon', methods=['POST'])
-def tryon_handler():
+def tryon():
+    data = request.json
+    human_image = process_image(data['human_image'])
+    garment_image = process_image(data['garment_image'])
+    description = data.get('description')
+    use_auto_mask = data.get('use_auto_mask', True)
+    use_auto_crop = data.get('use_auto_crop', False)
+    denoise_steps = int(data.get('denoise_steps', 30))
+    seed = int(data.get('seed', 42))
+    categorie = data.get('categorie' , 'upper_body')
+    human_dict = {
+        'background': human_image,
+        'layers': [human_image] if not use_auto_mask else None,
+        'composite': None
+    }
+    #clear_gpu_memory()
+
+    output_image, mask_image = start_tryon(human_dict, garment_image, description, use_auto_mask, use_auto_crop, denoise_steps, seed , categorie)
+
+    output_base64 = encode_image_to_base64(output_image)
+    mask_base64 = encode_image_to_base64(mask_image)
+
+    return jsonify({
+        'output_image': output_base64,
+        'mask_image': mask_base64
+    })
+
+@app.route('/tryon-v2', methods=['POST'])
+def tryon_v2():
+
+    data = request.json
+    human_image_data = data['human_image']
+    garment_image_data = data['garment_image']
+
+    # Process images (base64 ou URL)
+    human_image = process_image(human_image_data)
+    garment_image = process_image(garment_image_data)
+
+    description = data.get('description')
+    use_auto_mask = data.get('use_auto_mask', True)
+    use_auto_crop = data.get('use_auto_crop', False)
+    denoise_steps = int(data.get('denoise_steps', 30))
+    seed = int(data.get('seed', random.randint(0, 9999999)))
+    categorie = data.get('categorie', 'upper_body')
+
+        # Vérifie si 'mask_image' est présent dans les données
+    mask_image = None
+    if 'mask_image' in data:
+        mask_image_data = data['mask_image']
+        mask_image = process_image(mask_image_data)
+    
+    human_dict = {
+        'background': human_image,
+        'layers': [mask_image] if not use_auto_mask else None,
+        'composite': None
+    }
+    output_image, mask_image = start_tryon(human_dict, garment_image, description, use_auto_mask, use_auto_crop, denoise_steps, seed , categorie)
+    return jsonify({
+        'image_id': save_image(output_image)
+    })
+
+@spaces.GPU 
+def generate_mask(human_img, categorie='upper_body'):
+    device = "cuda"
+    openpose_model.preprocessor.body_estimation.model.to(device)
+    pipe.to(device)
+
     try:
-        data = request.json
-        human_image = decode_image_from_base64(data['human_image'])
-        garment_image = decode_image_from_base64(data['garment_image'])
-        description = data.get('description')
-        use_auto_mask = data.get('use_auto_mask', True)
-        use_auto_crop = data.get('use_auto_crop', False)
-        denoise_steps = int(data.get('denoise_steps', 30))
-        seed = int(data.get('seed', 42))
-        category = data.get('category', 'upper_body')
-        
-        human_dict = {
-            'background': human_image,
-            'layers': [human_image] if not use_auto_mask else None,
-            'composite': None
-        }
-        clear_gpu_memory()
+        # Redimensionner l'image pour le modèle
+        human_img_resized = human_img.convert("RGB").resize((384, 512))
 
-        output_image, mask_image = start_tryon(
-            human_dict, garment_image, description, use_auto_mask, use_auto_crop, denoise_steps, seed, category
-        )
+        # Générer les points clés et le masque
+        keypoints = openpose_model(human_img_resized)
+        model_parse, _ = parsing_model(human_img_resized)
+        mask, _ = get_mask_location('hd', categorie, model_parse, keypoints)
 
-        output_base64 = encode_image_to_base64(output_image)
-        mask_base64 = encode_image_to_base64(mask_image)
+        # Redimensionner le masque à la taille d'origine de l'image
+        mask_resized = mask.resize(human_img.size)
+        
+        return mask_resized
+    except Exception as e:
+        logging.error(f"Error generating mask: {e}")
+        raise e
 
+   
+@app.route('/generate_mask', methods=['POST'])
+def generate_mask_api():
+    try:
+        # Récupérer les données de l'image à partir de la requête
+        data = request.json
+        base64_image = data.get('human_image')
+        categorie = data.get('categorie', 'upper_body')
+        
+        # Décodage de l'image à partir de base64
+        human_img = process_image(base64_image)
+        
+        # Appeler la fonction pour générer le masque
+        mask_resized = generate_mask(human_img, categorie)
+        
+        # Encodage du masque en base64 pour la réponse
+        mask_base64 = encode_image_to_base64(mask_resized)
+        
         return jsonify({
-            'output_image': output_base64,
             'mask_image': mask_base64
-        })
+        }), 200
     except Exception as e:
-        logging.error(f"Error in tryon_handler: {e}")
+        logging.error(f"Error generating mask: {e}")
         return jsonify({'error': str(e)}), 500
 
+# Route pour récupérer l'image générée
+@app.route('/api/get_image/<image_id>', methods=['GET'])
+def get_image(image_id):
+    # Construire le chemin complet de l'image
+    image_path = image_id  # Assurez-vous que le nom de fichier correspond à celui que vous avez utilisé lors de la sauvegarde
+
+    # Renvoyer l'image
+    try:
+        return send_file(image_path, mimetype='image/webp')
+    except FileNotFoundError:
+        return jsonify({'error': 'Image not found'}), 404
+
 if __name__ == "__main__":
-    load_models()  # Charge les modèles au démarrage
-    app.run(host='0.0.0.0', port=7860)
+    app.run(debug=False, host="0.0.0.0", port=7860)