app.py

import gradio as gr
import cv2, numpy as np
from PIL import Image

def uniform_lighting(image):
    # Convert PIL → OpenCV
    img = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)

    # 1️⃣ Convert to grayscale to estimate illumination
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

    # 2️⃣ Smooth illumination field (large Gaussian kernel)
    illumination = cv2.GaussianBlur(gray, (51, 51), 0)

    # 3️⃣ Compute normalization map using brightest region
    brightest = np.max(illumination)
    illum_norm = brightest / (illumination + 1e-5)
    illum_norm = np.repeat(illum_norm[:, :, np.newaxis], 3, axis=2)

    # 4️⃣ Apply correction
    corrected = np.clip(img.astype(np.float32) * illum_norm, 0, 255).astype(np.uint8)

    # Convert back to PIL for Gradio display
    corrected_rgb = cv2.cvtColor(corrected, cv2.COLOR_BGR2RGB)
    return Image.fromarray(corrected_rgb)

title = "🧵 Saree AI – Uniform Lighting Correction"
description = """
Upload a saree image with uneven lighting.
This tool will automatically detect illumination gradients and normalize brightness
using the brightest region as reference, giving you a flat, uniform look for catalog or dataset use.
"""

demo = gr.Interface(
    fn=uniform_lighting,
    inputs=gr.Image(label="Upload Saree Image", type="pil"),
    outputs=gr.Image(label="Corrected Image"),
    title=title,
    description=description,
    examples=[],
)

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