For now, indexing a 4D FrameBatch returns a (3D) Frame. We could instead return a 3D FrameBatch.
I suggest we first land https://github.com/pytorch/torchcodec/pull/284, then I'll submit a PR that makes the change of returning a 3D FrameBatch instead - it'll be easier to evaluate the UX changes that this involves.
Opening this issue so we don't forget to discuss this: https://github.com/pytorch/torchcodec/pull/283#discussion_r1815047315
For now, indexing a 4D FrameBatch returns a (3D) Frame. We could instead return a 3D FrameBatch.
I suggest we first land https://github.com/pytorch/torchcodec/pull/284, then I'll submit a PR that makes the change of returning a 3D FrameBatch instead - it'll be easier to evaluate the UX changes that this involves.
CC @ahmadsharif1