Running feature selection on the GPU

Hello everyone, I would appreciate your assistance on an issue.

We are trying to implement feature selection on images we are taking from a Realsense camera (the code is in Python). We are running the selection on a separate process.

So far we were using the OpenCV method GoodFeaturesToTrack, however in an attempt to improve the FPS of the application we want to move the selection to the GPU.

We tried using the OpenCV CUDA module and its function createGoodFeaturesToTrackDetector, however since we are using multiprocessing CUDA is giving us serious problems.

The only solution to that we could find was to “spawn” the process instead of forking it, but this slows down the runtime significantly and is overall inefficient.

My question is, do you happen to have some examples of how to implement CUDA with OpenCV, specifically do to feature selection?

Alternatively, is there perhaps a better way to do feature selection using an NVIDIA API instead of OpenCV, that integrates CUDA support?

Thank you for the help.