ddd

handler.py

@@ -112,7 +112,7 @@ class EndpointHandler:
 
         vr = VideoReader(video_path, ctx=cpu(0))
         total_frames = len(vr)
-        indices = self.get_index(total_frames, num_segments)
+        indices = get_index(total_frames, num_segments)
         frames = [Image.fromarray(vr[i].asnumpy()) for i in indices]
 
         transform = T.Compose([
@@ -123,6 +123,8 @@ class EndpointHandler:
         ])
         processed = [transform(frame) for frame in frames]  # each is [3, 224, 224]
         video_tensor = torch.stack(processed, dim=0)  # [T, 3, 224, 224]
+        video_tensor = video_tensor.unsqueeze(0)  # [1, T, 3, 224, 224]
+        video_tensor = video_tensor.permute(0, 2, 1, 3, 4)  # 💥 [1, 3, T, 224, 224]
         return video_tensor
 
     def get_index(self, num_frames: int, num_segments: int):