Update app.py

app.py

@@ -1,7 +1,6 @@
 import sys
 sys.stdout.reconfigure(line_buffering=True) 
 
-
 import os
 import numpy as np
 import requests
@@ -39,7 +38,8 @@ device = torch.device('cuda' if 'cuda' in device_name and torch.cuda.is_availabl
 print(f"🧠 Using device: {device}")
 
 print("...Loading Grounding DINO model...")
-gnd_model_id = "IDEA-Research/grounding-dino-tiny"
+# --- ⬇️ UPGRADED MODEL ⬇️ ---
+gnd_model_id = "IDEA-Research/grounding-dino-large"
 processor_gnd = AutoProcessor.from_pretrained(gnd_model_id)
 model_gnd = AutoModelForZeroShotObjectDetection.from_pretrained(gnd_model_id).to(device)
 
@@ -49,7 +49,8 @@ sam_model = sam_model_registry["vit_b"](checkpoint=sam_checkpoint).to(device)
 predictor = SamPredictor(sam_model)
 
 print("...Loading BGE model for text embeddings...")
-bge_model_id = "BAAI/bge-small-en-v1.5"
+# --- ⬇️ UPGRADED MODEL ⬇️ ---
+bge_model_id = "BAAI/bge-large-en-v1.5"
 tokenizer_text = AutoTokenizer.from_pretrained(bge_model_id)
 model_text = AutoModel.from_pretrained(bge_model_id).to(device)
 print("✅ All models loaded successfully.")
@@ -236,14 +237,14 @@ def compare_items():
                 # Text comparison is always done
                 text_emb_found = np.array(item['text_embedding'])
                 text_score = cosine_similarity(query_text_emb, text_emb_found)
-                print(f"     - Text Score: {text_score:.4f}")
+                print(f"        - Text Score: {text_score:.4f}")
 
                 # --- NEW: Check if BOTH items have visual features ---
                 has_query_image = 'shape_features' in query_item and query_item['shape_features']
                 has_item_image = 'shape_features' in item and item['shape_features']
 
                 if has_query_image and has_item_image:
-                    print("     - Both items have images. Performing visual comparison.")
+                    print("        - Both items have images. Performing visual comparison.")
                     # If both have images, proceed with full comparison
                     query_shape_feat = np.array(query_item['shape_features'])
                     query_color_feat = np.array(query_item['color_features']).astype("float32")
@@ -259,25 +260,25 @@ def compare_items():
                     texture_score = cv2.compareHist(query_texture_feat, found_texture, cv2.HISTCMP_CORREL)
 
                     raw_image_score = (FEATURE_WEIGHTS["shape"] * shape_score +
-                                   FEATURE_WEIGHTS["color"] * color_score +
-                                   FEATURE_WEIGHTS["texture"] * texture_score)
+                                     FEATURE_WEIGHTS["color"] * color_score +
+                                     FEATURE_WEIGHTS["texture"] * texture_score)
 
-                    print(f"     - Raw Image Score: {raw_image_score:.4f}")
+                    print(f"        - Raw Image Score: {raw_image_score:.4f}")
                     
                     image_score = stretch_image_score(raw_image_score)
                     
                     # Weighted average of image and text scores
                     final_score = 0.4 * image_score + 0.6 * text_score
-                    print(f"     - Image Score: {image_score:.4f} | Final Score: {final_score:.4f}")
+                    print(f"        - Image Score: {image_score:.4f} | Final Score: {final_score:.4f}")
                 
                 else:
                     # If one or both items lack an image, the final score is JUST the text score
-                    print("     - One or both items missing image. Using text score only.")
+                    print("        - One or both items missing image. Using text score only.")
                     final_score = text_score
 
                 # Check if the final score meets the threshold
                 if final_score >= FINAL_SCORE_THRESHOLD:
-                    print(f"     - ✅ ACCEPTED (Score >= {FINAL_SCORE_THRESHOLD})")
+                    print(f"        - ✅ ACCEPTED (Score >= {FINAL_SCORE_THRESHOLD})")
                     results.append({
                         "_id": item_id,
                         "score": round(final_score, 4),
@@ -286,7 +287,7 @@ def compare_items():
                         "objectImage": item.get("objectImage"),
                     })
                 else:
-                    print(f"     - ❌ REJECTED (Score < {FINAL_SCORE_THRESHOLD})")
+                    print(f"        - ❌ REJECTED (Score < {FINAL_SCORE_THRESHOLD})")
 
             except Exception as e:
                 print(f"  [Skipping] Item {item_id} due to processing error: {e}")