Create app.py

app.py

@@ -0,0 +1,114 @@
+import numpy as np
+import cv2
+import gradio as gr
+from skimage.morphology import convex_hull_image
+
+# =========================================================
+# Utility Functions
+# =========================================================
+
+def orient_to_portrait(img):
+    h, w = img.shape[:2]
+    if w > h:
+        img = cv2.rotate(img, cv2.ROTATE_90_CLOCKWISE)
+    return img
+
+def analyze_cv_only(image, threshold_pct, min_defect_size):
+    """
+    Simulates the ML pipeline using traditional image processing.
+    """
+    if image is None:
+        return None, "0.0%", "0.0%", None, None
+
+    # 1. Preprocessing
+    img_color = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
+    img_color = orient_to_portrait(img_color)
+    gray = cv2.cvtColor(img_color, cv2.COLOR_BGR2GRAY)
+    blurred = cv2.GaussianBlur(gray, (5, 5), 0)
+
+    # 2. Simulate Print Segmentation (Thresholding)
+    # Instead of a model, we use simple binary thresholding
+    val = int((threshold_pct / 100.0) * 255)
+    _, print_mask = cv2.threshold(blurred, val, 255, cv2.THRESH_BINARY_INV)
+
+    # Find largest contour to represent the print
+    contours, _ = cv2.findContours(print_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+    if not contours:
+        return image, "0.0%", "0.0%", None, None
+        
+    largest_cnt = max(contours, key=cv2.contourArea)
+    print_mask_clean = np.zeros_like(print_mask)
+    cv2.drawContours(print_mask_clean, [largest_cnt], -1, 255, thickness=-1)
+
+    # 3. Simulate Defect Detection (Edge Detection within the print)
+    # Detects high-contrast areas (cracks/voids) inside the print mask
+    edges = cv2.Canny(blurred, 50, 150)
+    defect_mask = cv2.bitwise_and(edges, print_mask_clean)
+    
+    # Filter defects by size
+    final_defect_mask = np.zeros_like(defect_mask)
+    def_cnts, _ = cv2.findContours(defect_mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+    for cnt in def_cnts:
+        if cv2.contourArea(cnt) >= min_defect_size:
+            cv2.drawContours(final_defect_mask, [cnt], -1, 255, thickness=-1)
+
+    # 4. Metrics
+    # Integrity: Ratio of print to its convex hull
+    try:
+        hull = convex_hull_image(print_mask_clean > 0)
+        integrity_score = np.sum(print_mask_clean > 0) / np.sum(hull)
+    except:
+        integrity_score = 0.0
+
+    # Quality: Inverse of defect area
+    print_area = np.sum(print_mask_clean > 0)
+    defect_area = np.sum(final_defect_mask > 0)
+    quality_score = 100.0 - ((defect_area / print_area) * 100.0) if print_area > 0 else 0.0
+
+    # 5. Visualization
+    vis = img_color.copy()
+    cv2.drawContours(vis, [largest_cnt], -1, (255, 0, 0), 2) # Blue outline
+    cv2.drawContours(vis, [cv2.convexHull(largest_cnt)], -1, (0, 0, 255), 2) # Red hull
+    
+    # Draw defects in green
+    vis[final_defect_mask > 0] = [0, 255, 0]
+
+    # Save outputs
+    mask_path = "combined_mask.png"
+    cv2.imwrite(mask_path, final_defect_mask)
+    vis_path = "analysis_output.png"
+    cv2.imwrite(vis_path, vis)
+
+    return cv2.cvtColor(vis, cv2.COLOR_BGR2RGB), f"{integrity_score*100:.1f}%", f"{quality_score:.1f}%", vis_path, mask_path
+
+# =========================================================
+# UI Layout
+# =========================================================
+custom_css = """
+.score-box { background-color: #f8f9fa; border-radius: 10px; padding: 15px; text-align: center; }
+.score-value { font-size: 2rem !important; font-weight: 700 !important; color: #2b6cb0 !important; }
+"""
+
+with gr.Blocks(css=custom_css) as app:
+    gr.Markdown("# CV-Only Print Analysis (No ML)")
+    
+    with gr.Row():
+        with gr.Column():
+            img_input = gr.Image(type="numpy", label="Input Image")
+            thresh_slider = gr.Slider(50, 255, value=127, label="Brightness Threshold")
+            size_slider = gr.Slider(0, 500, value=20, label="Min Defect Size")
+            run_btn = gr.Button("Analyze", variant="primary")
+            
+        with gr.Column():
+            img_output = gr.Image(label="Analysis Result")
+            with gr.Row():
+                area_out = gr.Markdown("0.0%", elem_classes=["score-value"])
+                surface_out = gr.Markdown("0.0%", elem_classes=["score-value"])
+
+    run_btn.click(
+        fn=analyze_cv_only,
+        inputs=[img_input, thresh_slider, size_slider],
+        outputs=[img_output, area_out, surface_out]
+    )
+
+app.queue().launch(server_name="0.0.0.0", server_port=7860, ssr_mode=False)