Loading Different size tensors


I am trying to load multiple tensors where each of them are either (3,64,64), (3,128,128) and (3,256,256) I want to use them without resizing, as an input to the segmentation network but for that, I will have to make my custom collate_fn. The issue with that is, I am unable to use the data in the form of a mini-batch as they are of different sizes. Is there any workaround for this ?.

I thought of making a data loader to load all 32x32 images in one iteration and 256x256 in the next iteration and so on. I’m not sure how to achieve this, please can some help me out with this? Thank you

Please, can someone help me out with this ? Thank you