diff --git a/video_swin_transformer.py b/video_swin_transformer.py
index d0af874..acb0d66 100644
--- a/video_swin_transformer.py
+++ b/video_swin_transformer.py
@@ -65,11 +65,14 @@ class VideoSwinTransformer(NNOperator):
             mean=[0.485, 0.456, 0.406],
             std=[0.229, 0.224, 0.225],
         )
+        self.model.eval()
 
     def decoder_video(self,  data: List[VideoFrame]):
         video = numpy.stack([img.astype(numpy.float32) / 255. for img in data], axis=0)
         assert len(video.shape) == 4
         video = video.transpose(3, 0, 1, 2)  # twhc -> ctwh
+        if self.skip_preprocess:
+            self.transform_cfgs.update(num_frames=None)
         video = transform_video(
             video=video,
             **self.transform_cfgs