Data augmentation in Dataloader batch

I want to implement dataloader below
how can I do this?

Could you explain the use case a bit more, please?
Would you like to randomly augment the sample/batch and concatenate these transformed versions to the current sample/batch?
If so, I assume the batch size would be variable, since these transformations are applied randomly.
In that case, I think the easiest way would be to apply the transformations inside the DataLoader loop and these transformed sample with the original batch to the new input.

Oh. such a good solution,
I can transform both image and label(Ground Truth)
and stack them and feed for network

Thank you so much!