does anyone here know why the opencv 4.1.1 included in Deeptream (4 and 5) do not have CUDA support?
You may find some info here.
In short, build it yourself. You may use this script (should support all recent models ofJeston):
Ok, i thought there was some reason like optimizing GPU workload behind that. Seems it is about quality management because of the vast functionality of openCV. Thanks for pointing me to this. We have already build openCV with CUDA but i wondered if it would be wise to use it.
Be aware that there is no automatic backend selection in opencv.
For using GPU on Jetson, you would have to use opencv_cuda modules, using GpuMat instead of cv::Mat.
Only a subset of opencv algorithms are supported with opencv_cuda modules, and sometimes the results may be different.
Support for opencv cuda programming might be be limited in python.