Update app.py

app.py

@@ -4,7 +4,7 @@ from PIL import Image
 from io import BytesIO
 import torch
 import base64
-import io
+import io 
 import logging
 import gradio as gr
 import numpy as np
@@ -33,14 +33,12 @@ app = Flask(__name__)
 base_path = 'yisol/IDM-VTON'
 example_path = os.path.join(os.path.dirname(__file__), 'example')
 
-# Load models
 unet = UNet2DConditionModel.from_pretrained(
     base_path,
     subfolder="unet",
     torch_dtype=torch.float16,
 )
 unet.requires_grad_(False)
-
 tokenizer_one = AutoTokenizer.from_pretrained(
     base_path,
     subfolder="tokenizer",
@@ -84,33 +82,31 @@ UNet_Encoder = UNet2DConditionModel_ref.from_pretrained(
 parsing_model = Parsing(0)
 openpose_model = OpenPose(0)
 
-# Set models to evaluation mode
 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]),
-    ]
-)
+            [
+                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,
 )
 pipe.unet_encoder = UNet_Encoder
 
@@ -119,11 +115,15 @@ def pil_to_binary_mask(pil_image, threshold=0):
     grayscale_image = Image.fromarray(np_image).convert("L")
     binary_mask = np.array(grayscale_image) > threshold
     mask = np.zeros(binary_mask.shape, dtype=np.uint8)
-    mask[binary_mask] = 1
+    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 decode_image_from_base64(base64_str):
     try:
         img_data = base64.b64decode(base64_str)
@@ -144,7 +144,7 @@ def encode_image_to_base64(img):
         raise
 
 @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)
@@ -169,93 +169,141 @@ def start_tryon(dict, garm_img, garment_des, is_checked, is_checked_crop, denois
 
     if is_checked:
         keypoints = openpose_model(human_img.resize((384, 512)))
-        model_parse, _ = parsing_model(human_img)
-        return model_parse
+        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(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)
 
-    # Generate mask for the garment
-    garment_mask = pil_to_binary_mask(garm_img)
-    garment_mask = garment_mask.resize((768, 1024))
-    garment_mask = torch.from_numpy(np.array(garment_mask)).float().unsqueeze(0).to(device) / 255.0
-    garment_mask = torch.cat([garment_mask, garment_mask, garment_mask], dim=1)
+    human_img_arg = _apply_exif_orientation(human_img.resize((384, 512)))
+    human_img_arg = convert_PIL_to_numpy(human_img_arg, format="BGR")
 
-    # Process human image and garment image
-    try:
-        img_human = tensor_transfrom(human_img)
-        img_garment = tensor_transfrom(garm_img)
-    except Exception as e:
-        logging.error(f"Error processing images: {e}")
-        raise
+    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))
 
-    # Generate image
-    try:
-        result = pipe(
-            image_human=img_human.unsqueeze(0).to(device),
-            image_garment=img_garment.unsqueeze(0).to(device),
-            garment_mask=garment_mask,
-            denoise_steps=denoise_steps,
-            seed=seed,
-        ).images[0]
-        return result
-    except Exception as e:
-        logging.error(f"Error generating image: {e}")
-        raise
+    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,
+                    )
 
-def combine_images_with_masks(img_top, img_bottom, mask_top, mask_bottom):
-    img_top_pil = Image.open(io.BytesIO(img_top)).convert("RGBA")
-    img_bottom_pil = Image.open(io.BytesIO(img_bottom)).convert("RGBA")
-    mask_top_pil = Image.open(io.BytesIO(mask_top)).convert("L")
-    mask_bottom_pil = Image.open(io.BytesIO(mask_bottom)).convert("L")
+                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]
 
-    img_top_pil = img_top_pil.resize((768, 1024))
-    img_bottom_pil = img_bottom_pil.resize((768, 1024))
-    mask_top_pil = mask_top_pil.resize((768, 1024))
-    mask_bottom_pil = mask_bottom_pil.resize((768, 1024))
+    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
 
-    img_top_pil.paste(img_bottom_pil, (0, 0), mask_bottom_pil)
-    img_top_pil.paste(img_top_pil, (0, 0), mask_top_pil)
 
-    with BytesIO() as output:
-        img_top_pil.save(output, format="PNG")
-        combined_img_str = base64.b64encode(output.getvalue()).decode("utf-8")
-    return combined_img_str
+def clear_gpu_memory():
+    torch.cuda.empty_cache()
+    torch.cuda.synchronize()
 
 @app.route('/tryon', methods=['POST'])
 def tryon():
+    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))
+    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
+    })
+
+    
+def combine_images_with_masks(tops_image, bottoms_image, mask, is_checked_crop, crop_size):
     try:
-        data = request.json
-        image_base64 = data.get('human_image')
-        garm_img_base64 = data.get('garment_image')
-        is_checked = data.get('is_checked', False)
-        is_checked_crop = data.get('is_checked_crop', False)
-        denoise_steps = data.get('denoise_steps', 20)
-        seed = data.get('seed', 42)
-        category = data.get('category', 'upper_body')
-
-        human_img = decode_image_from_base64(image_base64)
-        garm_img = decode_image_from_base64(garm_img_base64)
-
-        combined_image = start_tryon(
-            dict={"background": human_img},
-            garm_img=garm_img,
-            garment_des="",
-            is_checked=is_checked,
-            is_checked_crop=is_checked_crop,
-            denoise_steps=denoise_steps,
-            seed=seed,
-            categorie=category,
-        )
-        result = combine_images_with_masks(combined_image, mask)
-
-        return jsonify({
-            'image': result
-        })
+        # Logique de combinaison des images de haut et de bas
+        if is_checked_crop:
+            tops_image = tops_image.resize(crop_size)
+            bottoms_image = bottoms_image.resize(crop_size)
+            combined_image = Image.new('RGB', (tops_image.width, tops_image.height))
+            combined_image.paste(tops_image, (0, 0))
+            combined_image.paste(bottoms_image, (0, tops_image.height // 2))
+        else:
+            combined_image = Image.new('RGB', (tops_image.width, tops_image.height))
+            combined_image.paste(tops_image, (0, 0))
+            combined_image.paste(bottoms_image, (0, tops_image.height // 2))
+
+        return combined_image
+
     except Exception as e:
-        logging.error(f"Error in tryon route: {e}")
-        return jsonify({'error': str(e)}), 500
+        raise ValueError(f"Error combining images with masks: {e}")
 
-@app.route('/', methods=['GET'])
-def welcome():
-    return "Welcome to IDM-VTON API"
 
 if __name__ == "__main__":
     app.run(debug=True, host="0.0.0.0", port=7860)
+