I loaded the model once on the GPU, and would like to run the network on different cuda streams, meaning having two videos running simultaneously using same model. I managed to implement this, but can’t find where I need to change the cuda stream, for the Workspace to run on.
Maybe it will be more helpful for OpenCV I’m passing cv::cuda::Stream, is there a way that I can do the same with caffe2, while sharing the data or running the network? Also I’m using the C++ version not Python.
Thanks very much