Update app.py

app.py

@@ -1,11 +1,10 @@
 # --------------------------------------------------------------------------
-# UNIFIED AI SERVICE V3.1 (Corrected Segmentation & Logging)
+# UNIFIED AI SERVICE V3.2 (Debug Uploads & Refactored)
 # --------------------------------------------------------------------------
 # This service uses DINOv2 for image embeddings and BGE for text embeddings.
-# It performs intelligent filtering before scoring.
-# 1. Filters by object name, date, and location hierarchy.
-# 2. Extracts features using BGE (text) and DINOv2 (image).
-# 3. Scores items based on a hybrid of text and image similarity.
+# - Filtering is handled by the Node.js backend.
+# - For debugging, segmented images are uploaded to Uploadcare and the URL
+#   is printed to the console log.
 # --------------------------------------------------------------------------
 
 import sys
@@ -36,6 +35,12 @@ TEXT_FIELDS_TO_EMBED = ["brand", "material", "size", "colors"]
 SCORE_WEIGHTS = { "text_score": 0.4, "image_score": 0.6 }
 FINAL_SCORE_THRESHOLD = 0.5
 
+# --- Load Uploadcare Credentials from Environment Variables ---
+# Make sure to set this as a Secret in your Hugging Face Space settings.
+UPLOADCARE_PUBLIC_KEY = os.getenv('UPLOADCARE_PUBLIC_KEY')
+if not UPLOADCARE_PUBLIC_KEY:
+    print("⚠️ WARNING: UPLOADCARE_PUBLIC_KEY environment variable not set. Debug uploads will fail.")
+
 print("="*50)
 print("🚀 Initializing AI Service with DINOv2...")
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
@@ -99,7 +104,6 @@ def cosine_similarity(vec1, vec2):
 
 def segment_guided_object(image: Image.Image, object_label: str) -> Image.Image:
     prompt = f"a {object_label}."
-    
     print(f"  [Segment] Using simple prompt: '{prompt}'")
     image_rgb = image.convert("RGB")
     image_np = np.array(image_rgb)
@@ -130,13 +134,36 @@ def segment_guided_object(image: Image.Image, object_label: str) -> Image.Image:
     
     return Image.fromarray(segmented_np, 'RGB')
 
+def upload_to_uploadcare(image: Image.Image) -> str:
+    """Uploads a PIL Image to Uploadcare and returns the CDN URL."""
+    if not UPLOADCARE_PUBLIC_KEY:
+        return "UPLOADCARE_PUBLIC_KEY not configured."
+    
+    try:
+        # Convert PIL Image to in-memory bytes buffer
+        buffer = BytesIO()
+        image.save(buffer, format='PNG')
+        buffer.seek(0)
+        
+        files = { 'file': ('segmented_image.png', buffer, 'image/png') }
+        data = { 'UPLOADCARE_PUB_KEY': UPLOADCARE_PUBLIC_KEY, 'UPLOADCARE_STORE': '1' }
+        
+        response = requests.post('https://upload.uploadcare.com/base/', files=files, data=data)
+        response.raise_for_status()
+        
+        file_uuid = response.json().get('file')
+        cdn_url = f"https://ucarecdn.com/{file_uuid}/"
+        return cdn_url
+    except Exception as e:
+        return f"Uploadcare upload failed: {e}"
+
 # ==========================================================================
 # --- FLASK ENDPOINTS ---
 # ==========================================================================
 
 @app.route('/', methods=['GET'])
 def health_check():
-    return jsonify({"status": "Unified AI Service (DINOv2) is running"}), 200
+    return jsonify({"status": "Unified AI Service is running"}), 200
 
 @app.route('/process', methods=['POST'])
 def process_item():
@@ -162,6 +189,12 @@ def process_item():
                     image = Image.open(BytesIO(img_response.content))
                     
                     segmented_image = segment_guided_object(image, data['objectName'])
+                    
+                    # --- DEBUGGING STEP: Upload segmented image and log the URL ---
+                    debug_url = upload_to_uploadcare(segmented_image)
+                    print(f"    - 🐞 DEBUG URL: {debug_url}")
+                    # -----------------------------------------------------------
+
                     embedding = get_image_embedding(segmented_image)
                     image_embeddings.append(embedding)
                 except Exception as e:
@@ -185,7 +218,6 @@ def compare_items():
         search_list = payload['searchList']
         print(f"\n[COMPARE] Received {len(search_list)} pre-filtered candidates for '{query_item.get('objectName')}'.")
         
-        # --- SCORING ---
         results = []
         for item in search_list:
             item_id = item.get('_id')
@@ -207,18 +239,15 @@ def compare_items():
                 item_img_embs = item.get('image_embeddings', [])
                 if query_img_embs and item_img_embs:
                     all_img_scores = []
-                    print(f"      - Image Pair Scores:") # Header for detailed scores
+                    print(f"      - Image Pair Scores:")
                     for i, q_emb in enumerate(query_img_embs):
                         for j, i_emb in enumerate(item_img_embs):
-                            # Calculate the score for this specific pair
                             pair_score = cosine_similarity(q_emb, i_emb)
-                            # Print the score for this pair
                             print(f"        - Query Img {i+1} vs Item Img {j+1}: {pair_score:.4f}")
                             all_img_scores.append(pair_score)
                     if all_img_scores:
                         image_score = max(all_img_scores)
                 
-                # Use a clearer label for the max score
                 print(f"    - Max Image Score: {image_score:.4f}")
                 
                 # 3. Calculate Final Score