Is it possible to Process Tensors of Varrying Size on GPU

Can Pytorch data loader process Tensors of Varrying size on GPU,without padding ?

I think we will have collate issues .Only work around i got get is to process an image one by one on GPU using parallel cpu workers for Loading the images.

You can have your own collate function in the data loader that cats and returns sizes if you want. If that is preferable over padding/cropping is might be dubious.

Cat will throw exception .
I have not made customization to collate before .
If you are familiar could you an example pls .