Transforms.Resize([]) followed by Transforms.ToTensor()
vs
Transforms.ToTensor() followed by Transforms.Resize([])
I have tried both of these in a dataset in Transforms.Compose() and i find that the training and testing loss/accuracy changes drastically (87% to 94%) just by interchanging the transforms. Can someone tell me why and which one should be used, if there’s any preferred way to resize. I know resize works based on input (i.e.) PIL or Tensor and my tranforms are linked to that, the problem is which one should we use?