In my project I need to change the algorithm adaptiveThreshol of Opencv which work on CPU by a same algorithm which work on the GPU. I find in library Npp, this algorithm: nppiFilterThresholdAdaptiveBoxBorder_8u_C1R.
But when I compare these algorithms, the speed is better on the CPU that GPU.
I know, that adpativeThreshold use boxfilter to blur the image. When i apply nppiFilterBox_8u_C1R, i have same time that nppiFilterThresholdAdaptiveBoxBorder_8u_C1R. (CPU 0.05 s : GPU 0.25 s ).
My data is transfered before on GPU because i use unifed memory when i grab frame.
Did you have some information about these algorithms and why that need lot of time to give a result?
Or do you know an other algotihme that can give the same result thatcv::adaptiveThreshold .
Thanks you for your help.
If you need more information, tell me.