import torchvision , opencv cannot be imshow

[azuredsky]

🐛 Describe the bug

when import torchvision, the opencv cannot imshow, but when no import torchvision, imshow successed

import numpy as np
import cv2
import torch
import torchvision

if __name__ == '__main__':
        cap = cv2.VideoCapture(-1)
        assert cap.isOpened(), 'Cannot capture source'

        while cap.isOpened():

            ret, frame = cap.read()

            frame = cv2.flip(frame,1)
            cv2.imshow('mycreame',frame)
            if ( cv2.waitKey(30)  == 27 ):
                break
        cv2.destroyAllWindows()

Versions

Versions of relevant libraries:
[pip3] numpy==1.21.6
[pip3] torch==1.13.1
[pip3] torchaudio==0.13.1
[pip3] torchvision==0.14.1
[conda] blas 1.0 mkl http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
[conda] cpuonly 2.0 0 pytorch
[conda] mkl 2021.4.0 h06a4308_640 http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
[conda] mkl-service 2.4.0 py37h7f8727e_0 http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
[conda] mkl_fft 1.3.1 py37hd3c417c_0 http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
[conda] mkl_random 1.2.2 py37h51133e4_0 http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
[conda] numpy 1.21.6 pypi_0 pypi
[conda] numpy-base 1.21.5 py37ha15fc14_3 http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main
[conda] pytorch 1.13.1 py3.7_cpu_0 pytorch
[conda] pytorch-mutex 1.0 cpu pytorch
[conda] torchaudio 0.13.1 py37_cpu pytorch
[conda] torchvision 0.14.1 py37_cpu pytorch

[pip3] opencv-python 4.6.0.66

[ajithvcoder]

if you have pytorchvideo uninstall it and check , there is an issue already here #5940

[azuredsky]

I have not installed pytorchvideo， but I uninstall av ,then opencv will show success, I guess may the _video_backend of the torchvision ,av cause the hung

[pmeier]

@azuredsky as @ajithvallabai correctly points out, #5940 deals with this exact issue. It seems opencv and av installed from PyPI cannot coexist in the same environment. As shown in #5940 (comment), the only part that torchvision plays in that is the fact that we try to import av. Thus, I'm closing this as not actionable. If you need both libraries at the same time, I suggest you use a conda environment for virtualization and ensuring that both libraries use the same ffmpeg binary.