ggg

handler.py

@@ -53,7 +53,7 @@ class EndpointHandler:
 
         elif video_b64:
             video_bytes = base64.b64decode(video_b64)
-            pixel_values = self._load_video(video_bytes).unsqueeze(0)  # [1, T, 3, 224, 224]
+            pixel_values = self._load_video(video_bytes)
             if self.device == "cuda":
                 pixel_values = pixel_values.half()
             pixel_values = pixel_values.to(self.device)
@@ -121,10 +121,10 @@ class EndpointHandler:
             T.ToTensor(),
             T.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
         ])
-        processed = [transform(frame) for frame in frames]  # each is [3, 224, 224]
+        processed = [transform(frame) for frame in frames]  # each: [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]
+        video_tensor = video_tensor.permute(1, 0, 2, 3)  # [3, T, 224, 224]
+        video_tensor = video_tensor.unsqueeze(0)  # [1, 3, T, 224, 224] ✅
         return video_tensor
 
     def get_index(self, num_frames: int, num_segments: int):