Update app.py

app.py

@@ -23,24 +23,22 @@ def find_similarity(base64_image, text_input):
     # Preprocess the image
     image = preprocess(image).unsqueeze(0).to(device)
 
-    # Tokenize the text input
+    # Prepare input text
     text_tokens = clip.tokenize([text_input]).to(device)
 
     # Encode image and text features
-    with torch no grad():
-        image_features = model.encode_image(image)
-        text_features = model.encode_text(text_tokens)
 
-    # Calculate cosine similarity
-    similarity = (image_features @ text_features.T).squeeze(0).cpu().numpy()
 
-    # Convert each element in the similarity array to Decimal
-    similarity_decimal = [Decimal(float(score)) for score in similarity]
+    with torch.no_grad():
+        image_features = model.encode_image(image)
+        text_features = model.encode_text(text_tokens)
 
-    # Format Decimal values as floats with specific precision (e.g., 4 decimal places)
-    formatted_similarity = [f'{float(score):.5f}' for score in similarity_decimal]
+    # Normalize features and calculate similarity
+    image_features /= image_features.norm(dim=-1, keepdim=True)
+    text_features /= text_features.norm(dim=-1, keepdim=True)
+    similarity = (text_features @ image_features.T).squeeze(0).cpu().numpy()
 
-    return formatted_similarity
+    return similarity[0, 0]
 
 # Create a Gradio interface
 iface = gr.Interface(