Dataloader missing temporal dimension?

I'm using the Charades dataset class in charadesrgb.py to construct a Dataset object and then feeding it into a Dataloader. However, it seems like the mini-batches returned by my Dataloader are of dimension BxCxHxW (missing the temporal T dimension).

Is this the expected behavior? Or am I using this incorrectly?

Start with charadesrgb.py and inspect how the Charades dataset constructs samples before they are passed to the DataLoader. Reproduce the reported BxCxHxW batch shape and determine whether the dataset or batching behavior is responsible for the missing temporal dimension; the expected shape is not specified in the issue.

Written by the indexing model from the issue text.