app.py

import gradio as gr
import numpy as np
import torch
import cv2
from PIL import Image
from torchvision import transforms
from cloth_segmentation.networks.u2net import U2NET  

# Load U²-Net model
model_path = "cloth_segmentation/networks/u2net.pth"
model = U2NET(3, 1)
state_dict = torch.load(model_path, map_location=torch.device('cpu'))
state_dict = {k.replace('module.', ''): v for k, v in state_dict.items()}
model.load_state_dict(state_dict)
model.eval()

def refine_mask(mask):
    """Refines mask using morphological closing followed by Gaussian blur"""
    kernel = np.ones((7, 7), np.uint8)
    mask = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel)  # Close holes inside dress
    mask = cv2.GaussianBlur(mask, (7, 7), 1.5)
    return mask

def segment_dress(image_np):
    """Segment dress using U²-Net"""
    transform_pipeline = transforms.Compose([
        transforms.ToTensor(),
        transforms.Resize((320, 320))
    ])

    image = Image.fromarray(image_np).convert("RGB")
    input_tensor = transform_pipeline(image).unsqueeze(0)

    with torch.no_grad():
        output = model(input_tensor)[0][0].squeeze().cpu().numpy()

    output = (output - output.min()) / (output.max() - output.min() + 1e-8)  # Normalize to [0, 1]
    dress_mask = (output > 0.5).astype(np.uint8) * 255
    dress_mask = cv2.resize(dress_mask, (image_np.shape[1], image_np.shape[0]), interpolation=cv2.INTER_LINEAR)

    return refine_mask(dress_mask)

def apply_grabcut(image_np, dress_mask):
    """Refines the mask using GrabCut to avoid color bleeding"""
    bgd_model = np.zeros((1, 65), np.float64)
    fgd_model = np.zeros((1, 65), np.float64)

    mask = np.where(dress_mask > 0, cv2.GC_PR_FGD, cv2.GC_BGD).astype('uint8')

    # Get bounding box of the mask
    coords = cv2.findNonZero(dress_mask)
    x, y, w, h = cv2.boundingRect(coords)
    rect = (x, y, w, h)

    cv2.grabCut(image_np, mask, rect, bgd_model, fgd_model, 5, cv2.GC_INIT_WITH_MASK)

    refined_mask = np.where((mask == cv2.GC_FGD) | (mask == cv2.GC_PR_FGD), 255, 0).astype("uint8")
    return refine_mask(refined_mask)

def recolor_dress(image_np, dress_mask, target_color):
    """Changes dress color while keeping texture & lighting intact"""

    # Convert target color to LAB
    target_color_lab = cv2.cvtColor(np.uint8([[target_color]]), cv2.COLOR_BGR2LAB)[0][0]

    # Convert image to LAB
    img_lab = cv2.cvtColor(image_np, cv2.COLOR_RGB2LAB)

    # Compute mean LAB values of dress pixels
    dress_pixels = img_lab[dress_mask > 0]
    if len(dress_pixels) == 0:
        return image_np  # No dress detected

    mean_L, mean_A, mean_B = np.mean(dress_pixels, axis=0)

    # Apply LAB shift
    a_shift = target_color_lab[1] - mean_A
    b_shift = target_color_lab[2] - mean_B
    img_lab[..., 1] = np.clip(img_lab[..., 1] + (dress_mask / 255.0) * a_shift, 0, 255)
    img_lab[..., 2] = np.clip(img_lab[..., 2] + (dress_mask / 255.0) * b_shift, 0, 255)

    # Convert back to RGB
    img_recolored = cv2.cvtColor(img_lab.astype(np.uint8), cv2.COLOR_LAB2RGB)

    # Create feathered mask for smooth blending
    lightness_mask = (img_lab[..., 0] / 255.0)
    feathered_mask = cv2.GaussianBlur(dress_mask, (15, 15), 5)
    adaptive_feather = (feathered_mask * lightness_mask).astype(np.uint8)

    # Blend the recolored dress with the original image
    img_final = (image_np * (1 - adaptive_feather[..., None] / 255) + img_recolored * (adaptive_feather[..., None] / 255)).astype(np.uint8)

    return img_final

def change_dress_color(img, color):
    """Main function to change dress color naturally"""
    if img is None:
        return None

    img_np = np.array(img)

    # Get dress segmentation mask
    dress_mask = segment_dress(img_np)

    if dress_mask is None:
        return img  # No dress detected

    # Further refine mask with GrabCut
    dress_mask = apply_grabcut(img_np, dress_mask)

    # Convert the selected color to BGR
    color_map = {
        "Red": (0, 0, 255), "Blue": (255, 0, 0), "Green": (0, 255, 0), "Yellow": (0, 255, 255),
        "Purple": (128, 0, 128), "Orange": (0, 165, 255), "Cyan": (255, 255, 0), "Magenta": (255, 0, 255),
        "White": (255, 255, 255), "Black": (0, 0, 0)
    }
    new_color_bgr = np.array(color_map.get(color, (0, 0, 255)), dtype=np.uint8)

    # Apply recoloring with blending
    img_recolored = recolor_dress(img_np, dress_mask, new_color_bgr)

    return Image.fromarray(img_recolored)

# Gradio Interface
demo = gr.Interface(
    fn=change_dress_color,
    inputs=[
        gr.Image(type = "pil", label="Upload Dress Image"),
        gr.Radio(["Red", "Blue", "Green", "Yellow", "Purple", "Orange", "Cyan", "Magenta", "White", "Black"], label="Choose New Dress Color")
    ],
    outputs=gr.Image(type = "pil", label="Color Changed Dress"),
    title="AI-Powered Dress Color Changer",
    description="Upload an image of a dress and select a new color. The AI will change the dress color naturally while keeping the fabric texture."
)

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