Pytorch 1.6 torchvision.transforms.functional.rotate doens't accept tensor, only PIL

on github it seems the latest commit supports image of type tensor for rotation. also on documentation.
still when i run the command i get an error saying image must be PIL object.

i can bypass it by simply transforming to PIL but it causes trouble debugging another error occuring further down the chain.

so do i need to switch to nightly in order to get the rotate function to work with tensors?


Transform on tensors has been added to PyTorch 1.7, released few days ago, you need to update torchvision package to 0.8. Simply, update to latest stable build.