Update app.py

app.py

@@ -90,42 +90,45 @@ def extract_images_from_pptx(pptx_path):
 def get_text_embedding(text):
     return text_model.encode(text).tolist()
 
-# Extract Image Embeddings
+# Extract Image Embeddings and Reduce to 384 Dimensions
 def get_image_embedding(image_path):
     try:
         image = Image.open(image_path)
         inputs = processor(images=image, return_tensors="pt")
         with torch.no_grad():
             image_embedding = model.get_image_features(**inputs).numpy().flatten()
+        
+        # Ensure embedding is 384-dimensional
+        if len(image_embedding) != 384:
+            pca = PCA(n_components=384)
+            image_embedding = pca.fit_transform(image_embedding.reshape(1, -1)).flatten()
+        
         return image_embedding.tolist()
     except Exception as e:
         print(f"Error generating image embedding: {e}")
         return None
 
-# Reduce Embedding Dimensions (If Needed)
-def reduce_embedding_dim(embeddings):
-    try:
-        embeddings = np.array(embeddings)
-        n_components = min(embeddings.shape[0], embeddings.shape[1], 384)  # Ensure valid PCA size
-        pca = PCA(n_components=n_components)
-        return pca.fit_transform(embeddings).tolist()
-    except Exception as e:
-        print(f"Error in PCA transformation: {e}")
-        return embeddings.tolist()  # Return original embeddings if PCA fails
-
 # Store Data in ChromaDB
 def store_data(texts, image_paths):
     for i, text in enumerate(texts):
         if text:
-            collection.add(ids=[f"text_{i}"], embeddings=[get_text_embedding(text)], documents=[text])
-
+            text_embedding = get_text_embedding(text)
+            if len(text_embedding) == 384:
+                collection.add(ids=[f"text_{i}"], embeddings=[text_embedding], documents=[text])
+    
     all_embeddings = [get_image_embedding(img_path) for img_path in image_paths if get_image_embedding(img_path) is not None]
+    
     if all_embeddings:
         all_embeddings = np.array(all_embeddings)
-        transformed_embeddings = reduce_embedding_dim(all_embeddings) if all_embeddings.shape[0] > 1 else all_embeddings.tolist()
+        
+        # Apply PCA only if necessary
+        if all_embeddings.shape[1] != 384:
+            pca = PCA(n_components=384)
+            all_embeddings = pca.fit_transform(all_embeddings)
+        
         for j, img_path in enumerate(image_paths):
-            collection.add(ids=[f"image_{j}"], embeddings=[transformed_embeddings[j]], documents=[img_path])
-
+            collection.add(ids=[f"image_{j}"], embeddings=[all_embeddings[j].tolist()], documents=[img_path])
+    
     print("Data stored successfully!")
 
 # Process and Store from Files