Update app.py

app.py

@@ -15,8 +15,19 @@ model.eval()
 DATASET_DIR = Path("dataset")
 CACHE_FILE = "cache.pkl"
 
+# Define supported image formats
+IMAGE_EXTENSIONS = ["*.jpg", "*.jpeg", "*.png", "*.bmp", "*.gif", "*.webp", "*.tiff", "*.tif"]
+
+def get_all_image_files():
+    """Get all image files from dataset directory"""
+    image_files = []
+    for ext in IMAGE_EXTENSIONS:
+        image_files.extend(DATASET_DIR.glob(ext))
+        image_files.extend(DATASET_DIR.glob(ext.upper()))  # Also check uppercase
+    return image_files
+
 def get_embedding(image: Image.Image, device="cpu"):
-    # Use CLIP's built-in preprocessing instead of custom resize
+    # Use CLIP's built-in preprocessing
     inputs = processor(images=image, return_tensors="pt").to(device)
     model_device = model.to(device)
     with torch.no_grad():
@@ -25,9 +36,11 @@ def get_embedding(image: Image.Image, device="cpu"):
     emb = emb / emb.norm(p=2, dim=-1, keepdim=True)
     return emb
 
+@spaces.GPU
 def get_reference_embeddings():
     # Get all current image files
-    current_images = set(img_path.name for img_path in DATASET_DIR.glob("*.jpg"))
+    current_image_files = get_all_image_files()
+    current_images = set(img_path.name for img_path in current_image_files)
     
     # Load existing cache if it exists
     cached_embeddings = {}
@@ -44,11 +57,15 @@ def get_reference_embeddings():
         embeddings = {}
         device = "cuda" if torch.cuda.is_available() else "cpu"
         
-        for img_path in DATASET_DIR.glob("*.jpg"):
+        for img_path in current_image_files:
             print(f"Processing {img_path.name}...")
-            img = Image.open(img_path).convert("RGB")
-            emb = get_embedding(img, device=device)
-            embeddings[img_path.name] = emb.cpu()
+            try:
+                img = Image.open(img_path).convert("RGB")
+                emb = get_embedding(img, device=device)
+                embeddings[img_path.name] = emb.cpu()
+            except Exception as e:
+                print(f"Error processing {img_path.name}: {e}")
+                continue
         
         # Save updated cache
         with open(CACHE_FILE, "wb") as f:
@@ -94,15 +111,16 @@ def add_image(name: str, image):
     if not name.strip():
         return "Please provide a valid image name."
     
-    path = DATASET_DIR / f"{name}.jpg"
-    image.save(path)
+    # Save as PNG to preserve quality for all input formats
+    path = DATASET_DIR / f"{name}.png"
+    image.save(path, "PNG")
     
     # Use GPU for consistency if available
     device = "cuda" if torch.cuda.is_available() else "cpu"
     emb = get_embedding(image, device=device)
     
     # Add to current embeddings and save cache
-    reference_embeddings[f"{name}.jpg"] = emb.cpu()
+    reference_embeddings[f"{name}.png"] = emb.cpu()
     
     with open(CACHE_FILE, "wb") as f:
         pickle.dump(reference_embeddings, f)
@@ -120,4 +138,4 @@ add_interface = gr.Interface(fn=add_image,
                              allow_flagging="never")
 
 demo = gr.TabbedInterface([search_interface, add_interface], tab_names=["Search", "Add Product"])
-demo.launch()
+demo.launch(mcp_server=True)