Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import cv2 | |
| import numpy as np | |
| def reconstruct_shape(image): | |
| if image is None: | |
| return None | |
| # Check if the image has an alpha (transparency) channel | |
| if image.shape[2] == 4: | |
| mask = image[:, :, 3] | |
| color_img = image[:, :, :3] | |
| else: | |
| gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY) | |
| _, mask = cv2.threshold(gray, 10, 255, cv2.THRESH_BINARY) | |
| color_img = image | |
| # Find contours based on the mask | |
| contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) | |
| if not contours: | |
| return image | |
| # Find the largest contour (the main object) | |
| c = max(contours, key=cv2.contourArea) | |
| # Extract the average color of the original shape for the reconstructed parts | |
| mean_val = cv2.mean(color_img, mask=mask) | |
| # Make the reconstructed part slightly distinct, or keep it exactly the average color | |
| dominant_color = (int(mean_val[0]), int(mean_val[1]), int(mean_val[2]), 255) | |
| # --- Shape Classification --- | |
| x, y, w, h = cv2.boundingRect(c) | |
| rect_area = w * h | |
| (cx, cy), radius = cv2.minEnclosingCircle(c) | |
| circle_area = np.pi * (radius ** 2) | |
| contour_area = cv2.contourArea(c) | |
| rect_ratio = contour_area / rect_area if rect_area > 0 else 0 | |
| circle_ratio = contour_area / circle_area if circle_area > 0 else 0 | |
| # Create a blank transparent canvas for the output | |
| output_image = np.zeros((image.shape[0], image.shape[1], 4), dtype=np.uint8) | |
| # --- Step 1: Draw the Reconstructed Base Shape --- | |
| if circle_ratio > rect_ratio: | |
| center = (int(cx), int(cy)) | |
| cv2.circle(output_image, center, int(radius), dominant_color, -1) | |
| else: | |
| if 0.85 <= w / h <= 1.15: | |
| side = max(w, h) | |
| cv2.rectangle(output_image, (x, y), (x + side, y + side), dominant_color, -1) | |
| else: | |
| cv2.rectangle(output_image, (x, y), (x + w, y + h), dominant_color, -1) | |
| # --- Step 2: Paste the Original Image Over the Base --- | |
| # This keeps the original texture intact while the missing parts show the dominant color | |
| original_area = mask > 0 | |
| output_image[original_area] = image[original_area] | |
| # --- Step 3: Draw the "Crack" / Seam --- | |
| # Draw a dark line along the boundary of the original shape to show where it was reconstructed | |
| crack_color = (40, 40, 40, 255) # Dark gray/black line for the crack | |
| crack_thickness = 3 # Adjust thickness of the line here | |
| cv2.drawContours(output_image, [c], -1, crack_color, crack_thickness) | |
| return output_image | |
| # --- Gradio Interface Setup --- | |
| with gr.Blocks(theme=gr.themes.Soft()) as demo: | |
| gr.Markdown("# 🔷 Irregular Shape Reconstructor (With Visible Seams)") | |
| gr.Markdown("Upload an image of an irregular shape. The algorithm reconstructs its original geometric form and draws a visible 'crack' line to show exactly where the new parts were glued on.") | |
| with gr.Row(): | |
| with gr.Column(): | |
| input_image = gr.Image(label="Upload Irregular Shape", image_mode="RGBA", type="numpy") | |
| submit_btn = gr.Button("Reconstruct Shape", variant="primary") | |
| with gr.Column(): | |
| output_image = gr.Image(label="Reconstructed Shape with Crack", image_mode="RGBA") | |
| submit_btn.click(fn=reconstruct_shape, inputs=input_image, outputs=output_image) | |
| if __name__ == "__main__": | |
| demo.launch() |