app.py

import os
import cv2
import gradio as gr
import mediapipe as mp
import numpy as np
from PIL import Image
from gradio_client import Client, handle_file

# Set up paths
example_path = os.path.join(os.path.dirname(__file__), 'example')

garm_list = os.listdir(os.path.join(example_path, "cloth"))
garm_list_path = [os.path.join(example_path, "cloth", garm) for garm in garm_list]

human_list = os.listdir(os.path.join(example_path, "human"))
human_list_path = [os.path.join(example_path, "human", human) for human in human_list]

# Initialize MediaPipe Pose
mp_pose = mp.solutions.pose
pose = mp_pose.Pose(static_image_mode=True)
mp_drawing = mp.solutions.drawing_utils
mp_pose_landmark = mp_pose.PoseLandmark

def detect_pose(image):
    # Convert to RGB
    image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

    # Run pose detection
    result = pose.process(image_rgb)

    keypoints = {}

    if result.pose_landmarks:
        # Draw landmarks on image
        mp_drawing.draw_landmarks(image, result.pose_landmarks, mp_pose.POSE_CONNECTIONS)

        # Get image dimensions
        height, width, _ = image.shape

        # Extract specific landmarks
        landmark_indices = {
            'left_shoulder': mp_pose_landmark.LEFT_SHOULDER,
            'right_shoulder': mp_pose_landmark.RIGHT_SHOULDER,
            'left_hip': mp_pose_landmark.LEFT_HIP,
            'right_hip': mp_pose_landmark.RIGHT_HIP
        }

        for name, index in landmark_indices.items():
            lm = result.pose_landmarks.landmark[index]
            x, y = int(lm.x * width), int(lm.y * height)
            keypoints[name] = (x, y)

            # Draw a circle + label for debug
            cv2.circle(image, (x, y), 5, (0, 255, 0), -1)
            cv2.putText(image, name, (x + 5, y - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 255, 255), 1)

    return image

def align_clothing(body_img, clothing_img):
    image_rgb = cv2.cvtColor(body_img, cv2.COLOR_BGR2RGB)
    result = pose.process(image_rgb)
    output = body_img.copy()

    if result.pose_landmarks:
        h, w, _ = output.shape

        # Extract key points
        def get_point(landmark_id):
            lm = result.pose_landmarks.landmark[landmark_id]
            return int(lm.x * w), int(lm.y * h)

        left_shoulder = get_point(mp_pose_landmark.LEFT_SHOULDER)
        right_shoulder = get_point(mp_pose_landmark.RIGHT_SHOULDER)
        left_hip = get_point(mp_pose_landmark.LEFT_HIP)
        right_hip = get_point(mp_pose_landmark.RIGHT_HIP)

        # Destination box (torso region)
        dst_pts = np.array([
            left_shoulder,
            right_shoulder,
            right_hip,
            left_hip
        ], dtype=np.float32)

        # Source box (clothing image corners)
        src_h, src_w = clothing_img.shape[:2]
        src_pts = np.array([
            [0, 0],
            [src_w, 0],
            [src_w, src_h],
            [0, src_h]
        ], dtype=np.float32)

        # Compute perspective transform and warp
        matrix = cv2.getPerspectiveTransform(src_pts, dst_pts)
        warped_clothing = cv2.warpPerspective(clothing_img, matrix, (w, h), borderMode=cv2.BORDER_TRANSPARENT)

        # Handle transparency
        if clothing_img.shape[2] == 4:
            alpha = warped_clothing[:, :, 3] / 255.0
            for c in range(3):
                output[:, :, c] = (1 - alpha) * output[:, :, c] + alpha * warped_clothing[:, :, c]
        else:
            output = cv2.addWeighted(output, 0.8, warped_clothing, 0.5, 0)

    return output

def process_image(human_img_path, garm_img_path):
    client = Client("franciszzj/Leffa")

    result = client.predict(
        src_image_path=handle_file(human_img_path),
        ref_image_path=handle_file(garm_img_path),
        ref_acceleration=False,
        step=30,
        scale=2.5,
        seed=42,
        vt_model_type="viton_hd",
        vt_garment_type="upper_body",
        vt_repaint=False,
        api_name="/leffa_predict_vt"
    )

    print(result)
    generated_image_path = result[0]
    print("generated_image_path" + generated_image_path)
    generated_image = Image.open(generated_image_path)

    return generated_image

# Custom CSS for better styling
custom_css = """
.gradio-container {
    max-width: 1200px !important;
}
.container {
    max-width: 1200px;
    margin: auto;
    padding: 20px;
}
.header {
    text-align: center;
    margin-bottom: 30px;
}
.header h1 {
    font-size: 2.5rem;
    margin-bottom: 10px;
    background: linear-gradient(45deg, #FF6B6B, #4ECDC4);
    -webkit-background-clip: text;
    -webkit-text-fill-color: transparent;
}
.header p {
    font-size: 1.1rem;
    color: #666;
}
.image-container {
    border-radius: 10px;
    overflow: hidden;
    box-shadow: 0 4px 8px rgba(0,0,0,0.1);
}
.upload-section {
    background: #f9f9f9;
    padding: 20px;
    border-radius: 10px;
    margin-bottom: 20px;
}
.try-btn {
    background: linear-gradient(45deg, #FF6B6B, #4ECDC4) !important;
    color: white !important;
    font-weight: bold !important;
    padding: 12px 24px !important;
    border-radius: 50px !important;
    border: none !important;
}
.try-btn:hover {
    transform: translateY(-2px);
    box-shadow: 0 6px 12px rgba(0,0,0,0.15);
}
.examples-section {
    margin-top: 15px;
}
.examples-section h3 {
    margin-bottom: 10px;
    color: #555;
}
"""

with gr.Blocks(css=custom_css, title="Virtual Try-On Fashion") as demo:
    with gr.Column(elem_classes=["container"]):
        with gr.Column(elem_classes=["header"]):
            gr.HTML("""
            <h1>Virtual Try-On Fashion</h1>
            <p>Upload your photo and select a garment to see how it looks on you! ✨</p>
            """)
        
        with gr.Row():
            with gr.Column(elem_classes=["upload-section"]):
                gr.Markdown("### Step 1: Upload Your Photo")
                human_img = gr.Image(
                    type="filepath", 
                    label='Person Image',
                    interactive=True,
                    elem_classes=["image-container"]
                )
                with gr.Column(elem_classes=["examples-section"]):
                    gr.Markdown("**Example poses:**")
                    example = gr.Examples(
                        inputs=human_img,
                        examples_per_page=5,
                        examples=human_list_path,
                        label=None
                    )
            
            with gr.Column(elem_classes=["upload-section"]):
                gr.Markdown("### Step 2: Select Garment")
                garm_img = gr.Image(
                    label="Clothing Item", 
                    type="filepath", 
                    interactive=True,
                    elem_classes=["image-container"]
                )
                with gr.Column(elem_classes=["examples-section"]):
                    gr.Markdown("**Example garments:**")
                    example = gr.Examples(
                        inputs=garm_img,
                        examples_per_page=5,
                        examples=garm_list_path,
                        label=None
                    )
            
            with gr.Column():
                gr.Markdown("### Step 3: See the Result")
                image_out = gr.Image(
                    label="Virtual Try-On Result", 
                    type="pil",
                    elem_classes=["image-container"],
                    interactive=False
                )
                with gr.Row():
                    gr.ClearButton([human_img, garm_img, image_out])
        
        with gr.Row():
            try_button = gr.Button(
                value="Try It On Now", 
                variant='primary',
                elem_classes=["try-btn"]
            )
        
        # Add some information sections
        with gr.Accordion("ℹ️ How to use this tool", open=False):
            gr.Markdown("""
            1. **Upload your photo**: Choose a clear front-facing photo with visible shoulders and hips
            2. **Select a garment**: Pick from our examples or upload your own clothing image
            3. **Click 'Try It On Now'**: See how the clothing looks on you instantly!
            
            For best results:
            - Use well-lit photos with good contrast
            - Avoid baggy clothing in your reference photo
            - Front-facing poses work best
            """)
        
        with gr.Accordion("⚠️ Limitations", open=False):
            gr.Markdown("""
            - Works best with upper body garments (shirts, jackets)
            - May not work perfectly with complex patterns or textures
            - Results depend on pose detection accuracy
            - Currently optimized for front-facing poses
            """)

    # Linking the button to the processing function
    try_button.click(
        fn=process_image, 
        inputs=[human_img, garm_img], 
        outputs=image_out,
        api_name="try_on"
    )

if __name__ == "__main__":
    demo.launch(show_error=True, share=False)