Update app.py

app.py

@@ -45,21 +45,6 @@ def load_image_from_url(url, max_retries=3):
             else:
                 return None
 
-def initialize_faiss_index(collection):
-    # 모든 임베딩을 가져와 numpy 배열로 변환
-    all_data = collection.get(include=['embeddings', 'metadatas'])
-    all_embeddings = np.array(all_data['embeddings']).astype('float32')
-    all_metadatas = all_data['metadatas']
-    
-    # faiss 인덱스 생성 및 임베딩 추가
-    dimension = all_embeddings.shape[1]
-    index = faiss.IndexFlatIP(dimension)  # 코사인 유사도를 사용하려면 IndexFlatIP를 사용
-    index.add(all_embeddings)
-    
-    return index, all_metadatas
-
-faiss_index, all_metadatas = initialize_faiss_index(collection)
-
 
 # 세그먼트 마스크 기반 임베딩 추출
 def get_segmented_embedding(img, final_mask):
@@ -170,15 +155,14 @@ def find_similar_images(query_embedding, collection, top_k=5):
     # 쿼리 임베딩 정규화 후 faiss 검색
     query_embedding = query_embedding.reshape(1, -1).astype('float32')
     faiss.normalize_L2(query_embedding)
-    _, indices = index.search(query_embedding, top_k)
+    distance, indices = index.search(query_embedding, top_k)
     
     # 검색된 상위 결과를 반환
     structured_results = []
     for metadata, idx in zip(all_metadatas, indices[0]):
-        similarity = np.dot(query_embedding, all_embeddings[idx]).item()  # 코사인 유사도 계산
         structured_results.append({
             'info': metadata,
-            'similarity': similarity
+            'similarity': 1-distance
         })
     
     return structured_results