app.py

import cv2
import mediapipe as mp
import numpy as np
from mediapipe.tasks import python
from mediapipe.tasks.python import vision
import gradio as gr

# Initialize MediaPipe hair segmentation model
MODEL_PATH = "hair_segmenter.tflite"
base_options = python.BaseOptions(model_asset_path=MODEL_PATH)
options = vision.ImageSegmenterOptions(
    base_options=base_options,
    running_mode=vision.RunningMode.IMAGE
)
segmenter = vision.ImageSegmenter.create_from_options(options)

def apply_hair_color(frame, mask, color, alpha=0.5, preserve_highlights=0.3):
    """Apply realistic hair color with preserved highlights"""
    lab = cv2.cvtColor(frame, cv2.COLOR_BGR2LAB)
    l, a, b = cv2.split(lab)
    highlight_mask = (l > 180).astype(np.float32) * preserve_highlights
    hair_mask = mask.astype(np.uint8)
    
    colored_hair = np.zeros_like(frame)
    colored_hair[:] = color[::-1]  # BGR format
    colored_lab = cv2.cvtColor(colored_hair, cv2.COLOR_BGR2LAB)
    colored_lab[:,:,0] = l * (1 + highlight_mask)
    
    blended_lab = cv2.addWeighted(lab, 1-alpha, colored_lab, alpha, 0)
    colored_result = cv2.cvtColor(blended_lab, cv2.COLOR_LAB2BGR)
    
    inverse_mask = cv2.bitwise_not(hair_mask)
    background = cv2.bitwise_and(frame, frame, mask=inverse_mask)
    hair_region = cv2.bitwise_and(colored_result, colored_result, mask=hair_mask)
    
    return cv2.add(background, hair_region)

def process_frame(frame, color_name, custom_color, intensity, highlights):
    """Process webcam frame with hair coloring"""
    # Professional hair color presets
    hair_colors = {
        "Chocolate Brown": [139, 69, 19],
        "Light Golden Brown": [222, 184, 135],
        "Soft Black": [51, 51, 51],
        "Golden Blonde": [218, 165, 32],
        "Dark Auburn": [139, 0, 0],
        "Deep Brown": [85, 46, 10],
        "Light Auburn": [205, 133, 63],
        "Medium Auburn": [165, 42, 42],
        "Platinum Blonde": [240, 220, 180],
        "Sienna Brown": [160, 82, 45]
    }
    
    # Get selected color
    if color_name == "Custom":
        # Convert hex to RGB
        selected_color = [int(custom_color[1:][i:i+2], 16) for i in (0, 2, 4)]
    else:
        selected_color = hair_colors[color_name]
    
    # Convert frame for MediaPipe
    frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
    mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=frame_rgb)
    
    # Perform hair segmentation
    segmentation_result = segmenter.segment(mp_image)
    
    if segmentation_result.confidence_masks:
        # Process mask
        mask = segmentation_result.confidence_masks[0].numpy_view()
        mask_resized = cv2.resize(mask, (frame.shape[1], frame.shape[0]))
        binary_mask = (mask_resized > 0.3).astype(np.uint8) * 255
        binary_mask = cv2.bitwise_not(binary_mask)
        
        # Clean up mask
        kernel = np.ones((5,5), np.uint8)
        binary_mask = cv2.erode(binary_mask, kernel, iterations=1)
        binary_mask = cv2.dilate(binary_mask, kernel, iterations=2)
        
        # Apply hair color
        colored_frame = apply_hair_color(
            frame,
            binary_mask,
            selected_color,
            alpha=intensity,
            preserve_highlights=highlights
        )
        return cv2.cvtColor(colored_frame, cv2.COLOR_BGR2RGB)
    
    return frame_rgb

def create_interface():
    """Create Gradio interface"""
    css = """
    .my-group {max-width: 800px !important; max-height: 800px !important;}
    .my-column {display: flex !important; justify-content: center !important; align-items: center !important;}
    """
    
    with gr.Blocks(css=css) as demo:
        gr.Markdown("# Virtual Hair Color Try-On")
        
        with gr.Row():
            with gr.Column():
                # Color selection controls
                color_dropdown = gr.Dropdown(
                    choices=["Chocolate Brown", "Light Golden Brown", "Soft Black", 
                            "Golden Blonde", "Dark Auburn", "Deep Brown", "Light Auburn",
                            "Medium Auburn", "Platinum Blonde", "Sienna Brown", "Custom"],
                    value="Chocolate Brown",
                    label="Choose Hair Color"
                )
                custom_color = gr.ColorPicker(
                    label="Custom Color",
                    visible=False
                )
                
                # Advanced controls
                intensity = gr.Slider(
                    minimum=0.0,
                    maximum=1.0,
                    value=0.8,
                    step=0.1,
                    label="Color Intensity"
                )
                highlights = gr.Slider(
                    minimum=0.0,
                    maximum=1.0,
                    value=0.3,
                    step=0.1,
                    label="Preserve Natural Highlights"
                )
            
            with gr.Column():
                # Modified WebRTC component for live preview
                webrtc = gr.Image(
                    label="Live Preview",
                    source="webcam",
                    streaming=True,
                    type="numpy"
                )
        
        # Show/hide custom color picker based on selection
        def update_color_picker(choice):
            return gr.ColorPicker(visible=(choice == "Custom"))
        
        color_dropdown.change(
            fn=update_color_picker,
            inputs=[color_dropdown],
            outputs=[custom_color]
        )
        
        # Stream processing with continuous updates
        webrtc.stream(
            fn=process_frame,
            inputs=[
                webrtc,
                color_dropdown,
                custom_color,
                intensity,
                highlights
            ],
            outputs=[webrtc],
            stream_every=0.1  # Process frame every 0.1 seconds for smooth updates
        )
    
    return demo

if __name__ == "__main__":
    demo = create_interface()
    demo.queue()
    demo.launch()