In pytorchvideo.data.labeled_video_dataset, is transform is wrongly defined? or am I using it wrongly?

[Tenglon]

In pytorchvideo.data.labeled_video_dataset,
transform is applied as:

class LabeledVideoDataset(torch.utils.data.IterableDataset):
    def __next__(self) -> dict:
        ....
        ....
            if self._transform is not None:

                sample_dict = self._transform(sample_dict) # This line is not correct?

where transform requires a dictionary as input,

But in pytorchvideo.transforms

Let's say, UniformTemporalSubsample, it takes a tensor as the input, which is inconsistent with a dictionary, am I missing something?

[Tenglon]

I figured out myself.

The transform should be defined as

        self._TRANSFORM = torchvision.transforms.Compose([
            pytorchvideo.transforms.ApplyTransformToKey(
                key="video",
                transform=torchvision.transforms.Compose([
                    pytorchvideo.transforms.UniformTemporalSubsample(self._CLIP_FRAMES),
                    pytorchvideo.transforms.Normalize((0.45, 0.45, 0.45), (0.225, 0.225, 0.225)),
                    pytorchvideo.transforms.RandomShortSideScale(min_size=256, max_size=320),
                    torchvision.transforms.RandomCrop(224),
                    torchvision.transforms.RandomHorizontalFlip(p=0.5),
                    ])
            )
        ])

, where ApplyTransformToKey takes a dictionary as input
so overall, it takes a dictionary as input.