app.py

app.py

@@ -50,14 +50,14 @@ def encode_search_query(search_query, model, device):
     with torch.no_grad():
         inputs = tokenizer([search_query],  padding=True, return_tensors="pt")
         #inputs = processor(text=[search_query], images=None, return_tensors="pt", padding=True)
-    text_features =  model.get_text_features(**inputs).detach().numpy()
+    text_features =  model.get_text_features(**inputs).cpu().numpy()
     return text_features
         
 # Find all matched photos
-def find_matches(text_features, photo_features, photo_ids, results_count=4):
+def find_matches(features, photo_ids, results_count=4):
   # Compute the similarity between the search query and each photo using the Cosine similarity
-  text_features = np.array(text_features)
-  similarities = (photo_features @ text_features.T).squeeze(1)
+  #text_features = np.array(text_features)
+  similarities = (photo_features @ features.T).squeeze(1)
   # Sort the photos by their similarity score
   best_photo_idx = (-similarities).argsort()
   # Return the photo IDs of the best matches
@@ -85,9 +85,9 @@ def image_search(search_text, search_image, option):
          processed_image = processor(text=None, images=search_image, return_tensors="pt", padding=True)["pixel_values"]
          image_feature = model.get_image_features(processed_image.to(device))
          image_feature /= image_feature.norm(dim=-1, keepdim=True)
-      image_feature = image_feature.detach().numpy()
+      image_feature = image_feature.cpu().numpy()
       # Find the matched Images
-      matched_images = find_matches(image_feature, photo_features, photo_ids, 4)
+      matched_images = find_matches(image_feature, photo_ids, 4)
       return show_output_image(matched_images)
   
 gr.Interface(fn=image_search,