Update pipeline/logic.py

pipeline/logic.py

@@ -24,26 +24,26 @@ def download_image_from_url(image_url: str) -> Image.Image:
     print(" [Download] ✅ Image downloaded and standardized to RGB.")
     return image_rgb
 
-# In pipeline/logic.py
-
 def detect_and_crop(image: Image.Image, object_name: str, models: dict) -> Image.Image:
     print(f"\n [Detect & Crop] Starting detection for object: '{object_name}'")
     image_np = np.array(image.convert("RGB"))
     height, width = image_np.shape[:2]
     prompt = [[f"a {object_name}"]]
     inputs = models['processor_gnd'](images=image, text=prompt, return_tensors="pt").to(models['device'])
+
     with torch.no_grad():
-        outputs = models['model_gnd'](**inputs)
+        outputs = models['model_gnd'](
+            **inputs,
+            box_threshold=0.4,
+            text_threshold=0.3
+        )
 
-    # THIS IS THE CORRECTED LINE:
     results = models['processor_gnd'].post_process_grounded_object_detection(
-        outputs,
-        inputs.input_ids,
-        object_detection_threshold=0.4, # Renamed from box_threshold
-        text_detection_threshold=0.3,   # Renamed from text_threshold
+        outputs=outputs,
+        input_ids=inputs.input_ids,
         target_sizes=[(height, width)]
     )
-    
+
     if not results or len(results[0]['boxes']) == 0:
         print(" [Detect & Crop] ⚠ Warning: Grounding DINO did not detect the object. Using full image.")
         return image
@@ -72,7 +72,15 @@ def extract_features(segmented_image: Image.Image) -> dict:
     mask = a
     gray = cv2.cvtColor(image_rgb, cv2.COLOR_BGR2GRAY)
     contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
-    hu_moments = cv2.HuMoments(cv2.moments(contours[0])).flatten() if contours else np.zeros(7)
+    if not contours:
+        # If no contours are found, return zero-filled features
+        print(" [Features] ⚠ Warning: No contours found in segmented image. Returning zero features.")
+        return {
+            "shape_features": [0.0] * 7,
+            "color_features": [0.0] * 512, # 8*8*8
+            "texture_features": [0.0] * 26
+        }
+    hu_moments = cv2.HuMoments(cv2.moments(contours[0])).flatten()
     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)
     color_hist = color_hist.flatten()
@@ -89,7 +97,7 @@ def extract_features(segmented_image: Image.Image) -> dict:
 
 def get_text_embedding(text: str, models: dict) -> list:
     print(f" [Embedding] Generating text embedding for: '{text[:50]}...'")
-    text_with_instruction = f"Represent this sentence for searching relevant passages: {text}"
+    text_with_instruction = f"Represent this description of a lost item for similarity search: {text}"
     inputs = models['tokenizer_text'](text_with_instruction, return_tensors='pt', padding=True, truncation=True, max_length=512).to(models['device'])
     with torch.no_grad():
         outputs = models['model_text'](**inputs)