Update app.py

app.py

@@ -1,10 +1,9 @@
 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 torch
 import base64
-import io 
 import logging
 import gradio as gr
 import numpy as np
@@ -24,24 +23,32 @@ from transformers import (
 from diffusers import DDPMScheduler, AutoencoderKL
 from utils_mask import get_mask_location
 from torchvision import transforms
-import apply_net
-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
+from torch.quantization import quantize_dynamic, default_dynamic_qconfig
+from torch.nn.utils import prune
 
 app = Flask(__name__)
 
 base_path = 'yisol/IDM-VTON'
 example_path = os.path.join(os.path.dirname(__file__), 'example')
 
+# Modèles avec quantization
 unet = UNet2DConditionModel.from_pretrained(
     base_path,
     subfolder="unet",
     torch_dtype=torch.float16,
     force_download=False
 )
+
+# Quantization dynamique des modèles pour une meilleure efficacité
+unet = quantize_dynamic(unet, {torch.nn.Linear}, dtype=torch.qint8)
+
 unet.requires_grad_(False)
+
+# Application de pruning pour réduire les poids inutiles
+for name, module in unet.named_modules():
+    if isinstance(module, torch.nn.Conv2d):
+        prune.l1_unstructured(module, name='weight', amount=0.2)
+
 tokenizer_one = AutoTokenizer.from_pretrained(
     base_path,
     subfolder="tokenizer",
@@ -70,12 +77,17 @@ text_encoder_two = CLIPTextModelWithProjection.from_pretrained(
     torch_dtype=torch.float16,
     force_download=False
 )
+
+# Autres modèles avec quantization
 image_encoder = CLIPVisionModelWithProjection.from_pretrained(
     base_path,
     subfolder="image_encoder",
     torch_dtype=torch.float16,
     force_download=False
 )
+
+image_encoder = quantize_dynamic(image_encoder, {torch.nn.Linear}, dtype=torch.qint8)
+
 vae = AutoencoderKL.from_pretrained(base_path,
                                     subfolder="vae",
                                     torch_dtype=torch.float16,
@@ -89,66 +101,46 @@ UNet_Encoder = UNet2DConditionModel_ref.from_pretrained(
     force_download=False
 )
 
-parsing_model = Parsing(0)
-openpose_model = OpenPose(0)
-
+# Désactivation de la mise à jour des poids
 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]),
-            ]
-    )
 
+# Pipeline Tryon
 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,
+        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
 
+tensor_transfrom = transforms.Compose([
+    transforms.ToTensor(),
+    transforms.Normalize([0.5], [0.5])
+])
+
+# Fonctions utilitaires optimisées
 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]:
-                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()  # 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
+    # Utilisation des opérations vectorisées pour améliorer les performances
+    grayscale_image = np.array(pil_image.convert("L")) > threshold
+    mask = (grayscale_image.astype(np.uint8) * 255)
+    return Image.fromarray(mask)
 
 def decode_image_from_base64(base64_str):
     try:
-        img_data = base64.b64decode(base64_str)
-        img = Image.open(BytesIO(img_data))
-        return img
+        return Image.open(BytesIO(base64.b64decode(base64_str)))
     except Exception as e:
         logging.error(f"Error decoding image: {e}")
         raise
@@ -157,136 +149,74 @@ def encode_image_to_base64(img):
     try:
         buffered = BytesIO()
         img.save(buffered, format="PNG")
-        img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
-        return img_str
+        return base64.b64encode(buffered.getvalue()).decode("utf-8")
     except Exception as e:
         logging.error(f"Error encoding image: {e}")
         raise
 
 def save_image(img):
-    unique_name = str(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
 
+# Optimisations du traitement de l'image avec GPU
 @spaces.GPU
-def start_tryon(dict, garm_img, garment_des, is_checked, is_checked_crop, denoise_steps, seed, categorie = '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)
-
     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((768, 1024))
-    else:
-        human_img = human_img_orig.resize((768, 1024))
+    human_img_orig = dict["background"].convert("RGB")
+    human_img = human_img_orig.resize((768, 1024))
 
     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_gray = get_mask_location('hd', categorie, model_parse, keypoints)
         mask = mask.resize((768, 1024))
     else:
         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'))
-    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:
-        return images[0], mask_gray
-
 
+            # Encodage des prompts
+            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
+            )
+
+            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 else None
+            images = pipe(
+                prompt_embeds=prompt_embeds,
+                negative_prompt_embeds=negative_prompt_embeds,
+                pooled_prompt_embeds=pooled_prompt_embeds,
+                negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+                num_inference_steps=denoise_steps,
+                generator=generator,
+                strength=1.0,
+                pose_img=pose_img,
+                cloth=garm_tensor,
+                mask_image=mask,
+                image=human_img,
+                height=1024,
+                width=768
+            )[0]
+
+    return images[0], mask_gray
+
+# Suppression explicite des caches GPU pour libérer la mémoire
 def clear_gpu_memory():
     torch.cuda.empty_cache()
     torch.cuda.synchronize()
 
-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():
     data = request.json
@@ -294,120 +224,27 @@ def tryon():
     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)
+    seed = int(data.get('seed', random.randint(0, 10000)))
 
     try:
-        # Redimensionner l'image pour le modèle
-        human_img_resized = human_img.convert("RGB").resize((384, 512))
-
-        # 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)
-
-        # 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
+        generated_image, _ = start_tryon(
+            human_image,
+            garment_image,
+            description,
+            use_auto_mask,
+            False,
+            denoise_steps,
+            seed
+        )
+        image_name = save_image(generated_image)
+        clear_gpu_memory()
 
-   
-@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({
-            'mask_image': mask_base64
-        }), 200
-    except Exception as 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
+            'generated_image': image_name
+        })
 
-    # Renvoyer l'image
-    try:
-        return send_file(image_path, mimetype='image/webp')
-    except FileNotFoundError:
-        return jsonify({'error': 'Image not found'}), 404
-
-if __name__ == "__main__":
-    app.run(debug=False, host="0.0.0.0", port=7860)
+    except Exception as e:
+        logging.error(f"Error during try-on: {e}")
+        clear_gpu_memory()
+        return jsonify({'error': 'Try-on failed.'}), 500