Is your feature request related to a problem? Please describe.
Currently, if using the default collate_fn, the interted data is a list of channel-first Tensor or numpy array, if we want to apply transforms on the data, most of the MONAI transforms can't support it.
For example, users want to use CopyItemsd, AsDiscreted, KeepLargestConnectedComponentd, etc.
All these transforms can only support a batch-first Tensor or a channel-first numpy array.

Thanks.