I could not get videocaptioning to work using the GPU and would always get a segmentation fault specifically on local cropsc = cv.resize{src=crop, dsize={256,256}}.

Doing some research on google I found someone else had a segementation fault using opencv because they were not using cudawarping and cuda to resize. I then noticed that when I luarocks torch-opencv it was warning me that it was not building with cuda support. I then found out that I did not install opencv with cuda support.

So after installing opencv with cuda support then installing torch-opencv with cuda support and lastly using cuda to resize, I was able to then use videocaptioning to work using the GPU. Tested working with OpenCV 3.1 and the latest torch-opencv.

I'm not sure if anyone is experiencing this issue and wondering if they were not how they got videocaptioning to work using the GPU.

I also tested the fix should continue to allow videocaptioning to work with the CPU instead of the GPU.