fix

app.py

@@ -27,8 +27,8 @@ FEATURE_WEIGHTS = {
 # ---- Load Models ----
 print("="*50)
 print("🚀 Initializing application and loading models...")
-# Set device to CPU for compatibility with Hugging Face Spaces free tier
-device = torch.device('cpu')
+device_name = os.environ.get("device", "cpu")
+device = torch.device('cuda' if 'cuda' in device_name and torch.cuda.is_available() else 'cpu')
 print(f"🧠 Using device: {device}")
 
 print("...Loading Grounding DINO model...")
@@ -118,12 +118,7 @@ def extract_features(segmented_image: Image.Image) -> dict:
     
     color_hist = cv2.calcHist([image_rgb], [0, 1, 2], mask, [8, 8, 8], [0, 256, 0, 256, 0, 256])
     cv2.normalize(color_hist, color_hist)
-    
-    # ------------------ THE FIX IS HERE ------------------
-    # The color_hist is multi-dimensional. We must flatten it to a 1D array
-    # before converting it to a list for the JSON response.
-    flat_color_hist = color_hist.flatten()
-    # ----------------------------------------------------
+    color_hist = color_hist.flatten()
     
     gray_masked = cv2.bitwise_and(gray, gray, mask=mask)
     lbp = feature.local_binary_pattern(gray_masked, P=24, R=3, method="uniform")
@@ -133,7 +128,7 @@ def extract_features(segmented_image: Image.Image) -> dict:
     
     return {
         "shape_features": hu_moments.tolist(),
-        "color_features": flat_color_hist.tolist(), # Use the flattened array
+        "color_features": color_hist.tolist(),
         "texture_features": texture_hist.tolist()
     }
 
@@ -168,11 +163,12 @@ def process_item():
 
         object_name = data.get('objectName')
         description = data.get('objectDescription')
-        image_url = data.get('objectImage') 
+        image_url = data.get('objectImage') # This can now be null
 
         if not all([object_name, description]):
             return jsonify({"error": "objectName and objectDescription are required."}), 400
 
+        # --- Always process text-based features ---
         canonical_label = get_canonical_label(object_name)
         text_embedding = get_text_embedding(description)
 
@@ -181,11 +177,13 @@ def process_item():
             "text_embedding": text_embedding,
         }
 
+        # --- Process visual features ONLY if an image_url is provided ---
         if image_url:
             print("--- Image URL provided, processing visual features... ---")
             image = download_image_from_url(image_url)
             object_crop = detect_and_crop(image, canonical_label)
             visual_features = extract_features(object_crop)
+            # Add visual features to the response
             response_data.update(visual_features)
         else:
             print("--- No image URL provided, skipping visual feature extraction. ---")
@@ -215,12 +213,9 @@ def compare_items():
             return jsonify({"error": "queryItem and searchList are required."}), 400
         
         query_text_emb = np.array(query_item['text_embedding'])
-        query_has_image = 'shape_features' in query_item and query_item['shape_features'] is not None
-
-        if query_has_image:
-            query_shape_feat = np.array(query_item['shape_features'])
-            query_color_feat = np.array(query_item['color_features']).astype("float32")
-            query_texture_feat = np.array(query_item['texture_features']).astype("float32")
+        query_shape_feat = np.array(query_item['shape_features'])
+        query_color_feat = np.array(query_item['color_features']).astype("float32")
+        query_texture_feat = np.array(query_item['texture_features']).astype("float32")
 
         results = []
         print(f"--- Comparing 1 query item against {len(search_list)} items ---")
@@ -233,29 +228,24 @@ def compare_items():
                 text_score = cosine_similarity(query_text_emb, text_emb_found)
                 print(f"     - Text Score: {text_score:.4f}")
 
-                final_score = text_score
-                image_score = 0.0
+                found_shape = np.array(item['shape_features'])
+                found_color = np.array(item['color_features']).astype("float32")
+                found_texture = np.array(item['texture_features']).astype("float32")
+
+                shape_dist = cv2.matchShapes(query_shape_feat, found_shape, cv2.CONTOURS_MATCH_I1, 0.0)
+                shape_score = 1.0 / (1.0 + shape_dist)
                 
-                item_has_image = 'shape_features' in item and item['shape_features'] is not None
-
-                if query_has_image and item_has_image:
-                    found_shape = np.array(item['shape_features'])
-                    found_color = np.array(item['color_features']).astype("float32")
-                    found_texture = np.array(item['texture_features']).astype("float32")
-
-                    shape_dist = cv2.matchShapes(query_shape_feat, found_shape, cv2.CONTOURS_MATCH_I1, 0.0)
-                    shape_score = 1.0 / (1.0 + shape_dist)
-                    
-                    color_score = cv2.compareHist(query_color_feat, found_color, cv2.HISTCMP_CORREL)
-                    texture_score = cv2.compareHist(query_texture_feat, found_texture, cv2.HISTCMP_CORREL)
-
-                    image_score = (FEATURE_WEIGHTS["shape"] * shape_score +
-                                   FEATURE_WEIGHTS["color"] * color_score +
-                                   FEATURE_WEIGHTS["texture"] * texture_score)
-                    
-                    final_score = 0.4 * image_score + 0.6 * text_score
-                    print(f"     - Image Score: {image_score:.4f} | Final Score: {final_score:.4f}")
+                color_score = cv2.compareHist(query_color_feat, found_color, cv2.HISTCMP_CORREL)
+                texture_score = cv2.compareHist(query_texture_feat, found_texture, cv2.HISTCMP_CORREL)
+
+                image_score = (FEATURE_WEIGHTS["shape"] * shape_score +
+                               FEATURE_WEIGHTS["color"] * color_score +
+                               FEATURE_WEIGHTS["texture"] * texture_score)
+                
+                final_score = 0.4 * image_score + 0.6 * text_score
                 
+                print(f"     - Image Score: {image_score:.4f} | Final Score: {final_score:.4f}")
+
                 results.append({
                     "_id": item_id,
                     "score": round(final_score, 4),
@@ -279,4 +269,4 @@ def compare_items():
         return jsonify({"error": str(e)}), 500
 
 if __name__ == '__main__':
-    app.run(host='0.0.0.0', port=7860)
+    app.run(host='0.0.0.0', port=7860)