traincascade speed and memory

asked 2017-05-30 07:08:37 -0500

CV_open gravatar image

I am using the traincascade.exe from OpenCV 3.1 and trying to speed up the processing time.

I set the precalcValBufSize and the precalcIdxBufSize both to 512 and then both to 2048. However the run time to create the classifier did not change from when no value was specified for these values (using default value of 1024), is there a problem with precalcValBufSize and the precalcIdxBufSize in OpenCV 3.1?

edit retag flag offensive close merge delete


Try to get the latest version. Just tested, and it seems that changing the parameters actually speeds up the process. However, it depends on how many data you are processing. IF your set is small enoguh to fit in 1024MB, then indeed raising it will not improve speed. I am assigning 10GB of RAM to each variable and when monitoring I see the application using all the assigned memory, and training goes tons faster, except for the sample grabbing part, which is still sequential.

StevenPuttemans gravatar imageStevenPuttemans ( 2017-05-31 04:13:27 -0500 )edit