Update app.py

app.py

@@ -59,7 +59,31 @@ ENDING = """For search acceleration capabilities, please refer to [Searchium.ai]
 
 
 DATA_PATH = '/home/user'
-downloaded_path = huggingface_hub.snapshot_download(repo_id="Searchium-ai/clip4clip-webvid150k", repo_type="model", cache_dir=DATA_PATH)
+# 1. 加载模型和处理器（只下载模型权重，约几百MB）
+model = CLIPModel.from_pretrained("Searchium-ai/clip4clip-webvid150k")
+processor = CLIPProcessor.from_pretrained("Searchium-ai/clip4clip-webvid150k")
+
+# 2. 将模型移到GPU（如果可用）或CPU
+device = "cuda" if torch.cuda.is_available() else "cpu"
+model = model.to(device)
+
+# 3. 如果需要提取文本嵌入，可以这样做：
+def get_text_embedding(text):
+    inputs = processor(text=text, return_tensors="pt", padding=True, truncation=True)
+    inputs = {k: v.to(device) for k, v in inputs.items()}
+    with torch.no_grad():
+        text_features = model.get_text_features(**inputs)
+    return text_features.cpu().numpy()
+
+# 4. 如果需要提取视频帧嵌入，可以这样做：
+def get_video_embedding(frames):
+    # frames: 一个PIL Image列表，每个元素是一帧
+    inputs = processor(images=frames, return_tensors="pt", padding=True)
+    inputs = {k: v.to(device) for k, v in inputs.items()}
+    with torch.no_grad():
+        video_features = model.get_image_features(**inputs)
+    # 对于多帧视频，可以对帧特征取平均
+    return video_features.mean(dim=0).cpu().numpy()
 
 ft_visual_features_file = downloaded_path +'/new_data/video_half_dataset_visual_features.npy'
 ft_visual_features_file_bin = downloaded_path + '/new_data/video_half_dataset_visual_features_binary_packed.npy'