Torchvision import breaks OpenCV code

🐛 Bug

When I import torchvisions in an opencv python code, the code crashes with “Segmentation Fault” (in console) or “Restarting kernel” in Spyder / iPython.

To Reproduce

Steps to reproduce the behavior:

1. Execute the code, the execution ends
2. Execute the code and uncomment the torchvision import, the execution gives “Segmentation fault”
3. Changing the order of import (torchvision then cv2) removes the issue for this short code sample but not for my complete code

import cv2
#import torchvision

pth = 'video.mp4'
cap = cv2.VideoCapture(pth)

ret,frame = cap.read()
cv2.imshow('frame',frame)

I also have an environment using Pytorch 1.5, the same code sample works fine under this environment. Thanks for checking if this is a real issue

Environment

PyTorch version: 1.8.0 Is debug build: False CUDA used to build PyTorch: 11.1 ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.1 LTS (x86_64) GCC version: (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0 Clang version: Could not collect CMake version: version 3.16.3 Libc version: glibc-2.10

Python version: 3.7 (64-bit runtime) Python platform: Linux-5.4.0-65-generic-x86_64-with-debian-bullseye-sid Is CUDA available: True CUDA runtime version: Could not collect

cuDNN version: Could not collect HIP runtime version: N/A MIOpen runtime version: N/A

Versions of relevant libraries: [pip3] numpy==1.18.1 [pip3] numpydoc==1.1.0 [pip3] torch==1.8.0 [pip3] torchaudio==0.8.0a0+a751e1d [pip3] torchfile==0.1.0 [pip3] torchlars==0.1.2 [pip3] torchsummary==1.5.1 [pip3] torchvision==0.9.0 [conda] blas 1.0 mkl
[conda] cudatoolkit 11.1.1 h6406543_8 conda-forge [conda] ffmpeg 4.3 hf484d3e_0 pytorch [conda] mkl 2020.2 256
[conda] mkl-service 2.3.0 py37he8ac12f_0
[conda] mkl_fft 1.2.0 py37h23d657b_0
[conda] mkl_random 1.1.1 py37h0573a6f_0
[conda] numpy 1.19.2 py37h54aff64_0
[conda] numpy-base 1.19.2 py37hfa32c7d_0
[conda] numpydoc 1.1.0 pyhd3eb1b0_1
[conda] pytorch 1.8.0 py3.7_cuda11.1_cudnn8.0.5_0 pytorch [conda] torch 1.7.1 pypi_0 pypi [conda] torchaudio 0.7.2 pypi_0 pypi [conda] torchfile 0.1.0 pypi_0 pypi [conda] torchlars 0.1.2 dev_0 <develop> [conda] torchvision 0.9.0 py37_cu111 pytorch

[andfoy]

Have you tried creating a clean environment that uses just conda-forge/defaults and not pip packages? That combination is always guaranteed to generate segfaults

[Whiax]

Yeah I saw the multiple versions, not sure if that could be the issue. I’m recreating a new clean environment to check that.

conda install torchvision -c pytorch conda install -c conda-forge opencv

If I do that, pytorch goes from 1.8.1 to 1.8.0 and torchvision also from 0.9.1 to 0.9.0. It installs opencv 4.5.2. And yeah ok, doing that solves the issue, so I guess it was on my side. Sorry for the trouble, it’s hard to get from where the bugs come from with these weird behaviors. _

For those with the same issue:

• Create a clean conda environment

Thanks for the help!

edit: I’ll also look on how to use spyder better, thanks for the tip