functional_tensor not found

[sdlpkxd]

when I run training code, it reports ModuleNotFoundError: No module named 'torchvision.transforms.functional_tensor'

my environment:
Mac M1 chip
python 3.9
torch 1.10.2
torchvision 0.2.2

Should torchvision.transforms.functional_tensor be torchvision.transforms.functional ?
Need help

[ssube]

I am also encountering this after upgrading to PyTorch 2.x. Based on checking a few venvs, it looks like torchvision.transforms.functional_tensor was available up through 0.14.1 but is not present in 0.15.0 or better.

[kamyker]

Works for me but i get:

The torchvision.transforms.functional_tensor module is deprecated in 0.15 and will be removed in 0.17. Please don't rely on it. You probably just need to use APIs in torchvision.transforms.functional or in torchvision.transforms.v2.functional.

[gordon0907]

I'm also having this warning. As I searched in the repo, torchvision.transforms.functional_tensor is only called in "basicsr/data/degradations.py". Replacing the import line torchvision.transforms.functional_tensor in this file, with torchvision.transforms.functional should solve the warning problem.

Environment:
macOS 13.4 (Apple M1 Max)
Python 3.10.11
torch 2.0.1
torchvision 0.15.2

[shun-lin]

Looks like this package is not updated anymore, to patch the issue in your python program you may add those lines

from torchvision.transforms import functional
import sys

sys.modules["torchvision.transforms.functional_tensor"] = functional

which will redirect torchvision.transforms.functional_tensor to torchvision.transforms.functional so the broken import will now work again w/ latest torchvision.

[sdlpkxd]

Yea, using torchvision.transforms.functional instread of torchvision.transforms.functional_tensor is recommended for the latest torchvision

[bedbad]

Why do users have to patch those every time?
This is definetely not solved. If the package is not updated anymore - change a maintaner or release managed fork
The package is used.

[lightandshadow68]

This is not a solution when BasicSR is used as a dependency in a library. Nor can I patch a file as part of an automated test suite.

[lightandshadow68]

I am also encountering this after upgrading to PyTorch 2.x. Based on checking a few venvs, it looks like torchvision.transforms.functional_tensor was available up through 0.14.1 but is not present in 0.15.0 or better.

It has been deprecated for some time now, and was eventually removed.