I installed Pytorch 1.0.0 using the wheel and torchvision using:
pip install torchvision. (it installs the latest version of torchvision)
The documentation (and the pointed code) of torchvision in the page of Pytorch 1.0.0 points to an old version of torchvision (this seems normal). However, it is not mentioned on the page what version of torchvision is!
My question is: is it safe to use the latest version of torchvision with old Pytorch version? if not, how to get the right torchvision version to a specific Pytorch version?
I came to this issue when I tried to override
transforms.RandomCrop using the code pointed in (doc). And I arrived to an inexplicable issue in inheritance… When I checked the code of
transforms.RandomCrop on my machine (with the latest torchvision) … I found out the reason for the inconsistency… that I am using a different
torchvision version … and the major difference in
For now, I downgraded torchvision to 0.2.1 (released by Apr 24, 2018) while Pytorch 1.0.0 was released by Dec 7, 2018. The next update of torchvision was by Feb. 27, 2019. I assume that the documentation in Pytorch 1.0.0 is pointing to torchvision 0.2.1.