Update app.py

app.py

@@ -1,93 +1,145 @@
 import os
+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
+import spaces
 import uuid
-import requests
-import torch
-from flask import Flask, request, jsonify, send_file
-from PIL import Image
-from io import BytesIO
-from torchvision import transforms
-from torchvision.transforms.functional import to_pil_image
+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
+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 src.tryon_pipeline import StableDiffusionXLInpaintPipeline as TryonPipeline
-from src.unet_hacked_garmnet import UNet2DConditionModel as UNet2DConditionModel_ref
-from src.unet_hacked_tryon import UNet2DConditionModel
-import apply_net
+from torchvision.transforms.functional import to_pil_image
 
 app = Flask(__name__)
 
 base_path = 'yisol/IDM-VTON'
 example_path = os.path.join(os.path.dirname(__file__), 'example')
 
-# Load models
-def load_model(name, subfolder, dtype=torch.float16):
-    return torch.load(
-        os.path.join(base_path, subfolder, name),
-        map_location=torch.device('cuda' if torch.cuda.is_available() else 'cpu'),
-        dtype=dtype
-    )
-
-unet = load_model("unet.pt", "unet")
-tokenizer_one = AutoTokenizer.from_pretrained(base_path, subfolder="tokenizer", use_fast=False)
-tokenizer_two = AutoTokenizer.from_pretrained(base_path, subfolder="tokenizer_2", use_fast=False)
+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)
-text_encoder_two = CLIPTextModelWithProjection.from_pretrained(base_path, subfolder="text_encoder_2", torch_dtype=torch.float16)
-image_encoder = CLIPVisionModelWithProjection.from_pretrained(base_path, subfolder="image_encoder", torch_dtype=torch.float16)
-vae = AutoencoderKL.from_pretrained(base_path, subfolder="vae", torch_dtype=torch.float16)
+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)
+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)
 
-# Disable gradient computation
-for model in [unet, UNet_Encoder, image_encoder, vae, text_encoder_one, text_encoder_two]:
-    model.requires_grad_(False)
-
-tensor_transfrom = transforms.Compose([
-    transforms.ToTensor(),
-    transforms.Normalize([0.5], [0.5]),
-])
+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
+        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"))
-    binary_mask = np_image > threshold
-    mask = (binary_mask * 255).astype(np.uint8)
-    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
@@ -95,7 +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
@@ -104,33 +157,36 @@ 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
 
 @spaces.GPU
-def start_tryon(human_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 = human_dict["background"].convert("RGB")
+    human_img_orig = dict["background"].convert("RGB")
 
     if is_checked_crop:
         width, height = human_img_orig.size
-        target_width = min(width, height * (3 / 4))
-        target_height = min(height, width * (4 / 3))
+        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
-        cropped_img = human_img_orig.crop((left, top, width - left, height - top))
+        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:
@@ -139,51 +195,78 @@ def start_tryon(human_dict, garm_img, garment_des, is_checked, is_checked_crop,
     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(human_dict['layers'][0].convert("RGB").resize((768, 1024)))
-    
+        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)[:, :, ::-1]
+    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(), torch.cuda.amp.autocast():
-        prompt = f"model is wearing {garment_des}"
-        negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
-        prompt_embeds = pipe.encode_prompt(prompt, num_images_per_prompt=1, do_classifier_free_guidance=True, negative_prompt=negative_prompt)
-        prompt = f"a photo of {garment_des}"
-        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=prompt_embeds[1].to(device, torch.float16),
-            pooled_prompt_embeds=prompt_embeds[2].to(device, torch.float16),
-            negative_pooled_prompt_embeds=prompt_embeds[3].to(device, torch.float16),
-            num_inference_steps=denoise_steps,
-            generator=generator,
-            strength=1.0,
-            pose_img=pose_img,
-            text_embeds_cloth=prompt_embeds_c.to(device, torch.float16),
-            cloth=garm_tensor,
-            mask_image=mask,
-            image=human_img,
-            height=1024,
-            width=768,
-            ip_adapter_image=garm_img.resize((768, 1024)),
-            guidance_scale=2.0
-        )[0]
+    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)
@@ -192,56 +275,126 @@ def start_tryon(human_dict, garm_img, garment_des, is_checked, is_checked_crop,
     else:
         return images[0], mask_gray
 
+
 def clear_gpu_memory():
     torch.cuda.empty_cache()
     torch.cuda.synchronize()
 
 def process_image(image_data):
-    if image_data.startswith(('http://', 'https://')):
-        return get_image_from_url(image_data)
-    return decode_image_from_base64(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
+    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:
-        human_image_data = process_image(data['human_image'])
-        garment_image_data = process_image(data['garment_image'])
-        category = data.get('category', 'upper_body')
-        description = data.get('description', '')
-        checked = data.get('checked', False)
-        checked_crop = data.get('checked_crop', False)
-        denoise_steps = data.get('denoise_steps', 50)
-        seed = data.get('seed', None)
-
-        human_dict = {
-            "background": human_image_data,
-            "layers": [human_image_data],
-        }
-
-        result_img, mask_img = start_tryon(
-            human_dict,
-            garment_image_data,
-            description,
-            checked,
-            checked_crop,
-            denoise_steps,
-            seed,
-            category
-        )
-
-        encoded_image = encode_image_to_base64(result_img)
-        encoded_mask = encode_image_to_base64(mask_img)
-
-        #clear_gpu_memory()
+        # 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
 
+   
+@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({
-            'result_image': encoded_image,
-            'mask_image': encoded_mask,
-        })
-
+            'mask_image': mask_base64
+        }), 200
     except Exception as e:
-        logging.error(f"Error in /tryon endpoint: {e}")
+        logging.error(f"Error generating mask: {e}")
         return jsonify({'error': str(e)}), 500
 
 # Route pour récupérer l'image générée
@@ -257,4 +410,4 @@ def get_image(image_id):
         return jsonify({'error': 'Image not found'}), 404
 
 if __name__ == "__main__":
-    app.run(debug=False, host="0.0.0.0", port=7860)
+    app.run(debug=False, host="0.0.0.0", port=7860)