How GPU (CUDA) HoG (Histogram of Oriented Gradients) and SVM classification is parallelized?

asked 2015-04-20 23:41:07 -0500

wolf112 gravatar image

updated 2015-04-22 03:18:33 -0500

Can someone please explain me how OpenCV GPU(CUDA) HOG and SVM classification works? How is it parallelized compared to CPU version ? I would like to know the process in terms of Cuda threads and kernels.

Thank you

edit retag flag offensive close merge delete

Comments

Offcourse it is parellelized and optimized, else it would not be in the GPU/CUDA folder. Basically what it does is parallelize the block processing for the histograms to be calculated. By doing multiple blocks in parallel speed increase can be enourmous. But to me this seems kind of obvious, since there is no way in parallelizing the histogram combining :D

StevenPuttemans gravatar imageStevenPuttemans ( 2015-04-21 02:36:41 -0500 )edit

Thank you for the reply, but how is the block processing is parallelized? I would like to know the process in terms of Cuda threads and kernels. I'm having a look at source file hog.cu but I cannot understand yet. I also want to know how the SVM classification is parallelized.

wolf112 gravatar imagewolf112 ( 2015-04-21 23:14:04 -0500 )edit

I haven't got a single clue :) I don't understand cuda code myself. Might need someone here with that experience...

StevenPuttemans gravatar imageStevenPuttemans ( 2015-04-22 03:12:08 -0500 )edit