OpenCv with neon coprocessor does not have increase in performace

asked 2014-04-25 00:15:05 -0500

surinder gravatar image

updated 2014-04-25 00:25:36 -0500

berak gravatar image

I have been working on Pandaboard opencv from a week to optimize OpenCv libraries to achieve frame rate of 30fps in Pandaboard OpenCv Ubuntu. but still no luck.

I compiled & built OpenCv 2.4.8 with enabled neon, TBB and libjpeg-turbo on Pandaboard for optimizing the performance of opencv.

I am using this tutorial to build OpenCv libraries.

After building the opencv It runs too slow and achieve a frame rate of 11 to 12 fps. Window size is 320x240.

Please help If you have faced this issue.

Thanks

edit retag flag offensive close merge delete

Comments

Your quote 'After building the opencv It runs too slow and achieve a frame rate of 11 to 12 fps. Window size is 320x240' is hardly saying anything useful. There is no average FPS for the entire OpenCV lib. So specify what you are doing, maybe the function you use hasn't had any neon optimizations yet ... Also running a window size of 320x240 on a panda board with 11 - 12 FPS if you apply heavy calculations is quite a good result...

StevenPuttemans gravatar imageStevenPuttemans ( 2014-04-25 03:13:03 -0500 )edit

yes, neon optimization is only for some functions. I am using highgui functions like cvSmooth, cvAbs. but cvSmooth takes lot of time to process a frame.

surinder gravatar imagesurinder ( 2014-04-28 04:51:41 -0500 )edit

First mistake, why use cvSmooth and not the GaussianBlur C++ function? You are using depricated C - API.

StevenPuttemans gravatar imageStevenPuttemans ( 2014-04-28 04:52:31 -0500 )edit