I have a batch of images in the usual tensor form (B, C, H, W). To apply the usual sets of rotations and flips, I’m using the construction here: Rotation transformation · Issue #566 · pytorch/vision · GitHub.
However, this isn’t quite doing what I want. torchvision.transforms — Torchvision 0.10.0 documentation states:
Randomized transformations will apply the same transformation to all the images of a given batch
I want a different random transform to be applied to each image in the batch, but I would still like to be able to process all the images together in a batch (avoiding an explicit loop). Is there any method available to do this?