ModuleNotFoundError: No module named 'torchvision.transforms.functional_tensor'

[rahulmahmud]

Please solve step by step
Traceback (most recent call last):
File "/content/Real-ESRGAN/inference_realesrgan_video.py", line 10, in
from basicsr.archs.rrdbnet_arch import RRDBNet
File "/usr/local/lib/python3.10/dist-packages/basicsr/init.py", line 4, in
from .data import *
File "/usr/local/lib/python3.10/dist-packages/basicsr/data/init.py", line 22, in
_dataset_modules = [importlib.import_module(f'basicsr.data.{file_name}') for file_name in dataset_filenames]
File "/usr/local/lib/python3.10/dist-packages/basicsr/data/init.py", line 22, in
_dataset_modules = [importlib.import_module(f'basicsr.data.{file_name}') for file_name in dataset_filenames]
File "/usr/lib/python3.10/importlib/init.py", line 126, in import_module
return _bootstrap._gcd_import(name[level:], package, level)
File "/usr/local/lib/python3.10/dist-packages/basicsr/data/realesrgan_dataset.py", line 11, in
from basicsr.data.degradations import circular_lowpass_kernel, random_mixed_kernels
File "/usr/local/lib/python3.10/dist-packages/basicsr/data/degradations.py", line 8, in
from torchvision.transforms.functional_tensor import rgb_to_grayscale
ModuleNotFoundError: No module named 'torchvision.transforms.functional_tensor'

[jundao17]

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.

[JossCamp]

you need to install a specific version of pytorch

python -m pip install torch==2.0.1 torchvision==0.15.2 --extra-index-url https://download.pytorch.org/whl/cu118

After that, it should work correctly.

[cncloudli]

This might be caused by the modification of this module name. I opened the same issue in BasicSR project. In some specific versions of pytorch, for this module name there is a "_" in front of "functional". That‘s why changing the version of pytorch could solve this issue.

[KanTakahiro]

Just because BasicSR doesn't get ready for torchvision >= 0.17
I found this and edited the basicsr/data/degradations.py file, then the issue solved.

[m4ra7h0n]

you need to install a specific version of pytorch

python -m pip install torch==2.0.1 torchvision==0.15.2 --extra-index-url https://download.pytorch.org/whl/cu118

After that, it should work correctly.

ERROR: Could not find a version that satisfies the requirement torch==2.0.1 (from versions: 2.2.0, 2.2.1, 2.2.2)
ERROR: No matching distribution found for torch==2.0.1

[DongKat]

@m4ra7h0n you should try older version of python.

[PhielAi]

C:\Users\ user-name \anaconda3\envs\ envs-name \Lib\site-packages\basicsr\data\degradations.py

line 8:
from torchvision.transforms.functional_tensor import rgb_to_grayscale
to:
from torchvision.transforms._functional_tensor import rgb_to_grayscale

[Anyeos]

It need to be updated because the AI world is continiously evolving and there are always new versions and solutions for everything. Now xformers requires torch==2.2.2 and it cannot be installed with torch==2.0.1
So the outdated implied library is the real problem.

[moustaphachadly]

C:\Users\ user-name \anaconda3\envs\ envs-name \Lib\site-packages\basicsr\data\degradations.py

line 8: from torchvision.transforms.functional_tensor import rgb_to_grayscale to: from torchvision.transforms._functional_tensor import rgb_to_grayscale

This worked for me, but with miniconda: /Users/[user-name]/miniconda3/lib/python3.12/site-packages/basicsr/data/degradations.py.

[wladradchenko]

Hi! Modifying the library code, as suggested in the answer by @PhielAi on Apr 9, is not ideal, and locking to specific versions can also introduce limitations. Instead, you can handle the missing function without changing the library code directly.

Here’s an approach that dynamically adds the rgb_to_grayscale function into the torchvision.transforms.functional_tensor module:

import sys
import types
from torchvision.transforms.functional import rgb_to_grayscale

# Create a module for `torchvision.transforms.functional_tensor`
functional_tensor = types.ModuleType("torchvision.transforms.functional_tensor")
functional_tensor.rgb_to_grayscale = rgb_to_grayscale

# Add this module to sys.modules so other imports can access it
sys.modules["torchvision.transforms.functional_tensor"] = functional_tensor

This approach makes the missing function available where it’s expected, without requiring any modifications to the library code itself. It's especially useful for introducing back functions that may have been removed in recent library updates, providing a clean and maintainable solution.