Bencmarking on Different OpenCV Builds

asked 2013-07-17 06:28:47 -0500

enienws gravatar image


I am planning to build OpenCV with different methods, such as with or without IPP, with gcc or icc etc.

I think on a benchmark application by using these different builds of OpenCV. But I cannot find a considerable way to handle this?

Could you please advice some techniques to use for this purpose?


edit retag flag offensive close merge delete


berak gravatar imageberak ( 2013-07-17 06:33:12 -0500 )edit

I think the benchmark idea is in fact pretty nice and usefull to the most of OpenCV developpers, only do not underestimate the time it will take to program it decently. I would go for basic computer vision techniques like thresholding, line detection, segmentation, feature detection, feature description, feature matching, ... however start with easy stuff, like just a plane canny or sobel edge detector.

StevenPuttemans gravatar imageStevenPuttemans ( 2013-07-17 07:00:40 -0500 )edit

berak, Shervin's post is valueable. But I am looking for something different. I think similiar to StevenPuttermans. I have built performance code when building OpenCV. Under the bin folder in build folder I have found some executables which are aimed to measure performance of different tasks. The following is the list of these executables:

opencv_perf_calib3d opencv_perf_nonfree opencv_perf_stitching opencv_perf_core opencv_perf_objdetect opencv_perf_superres opencv_perf_features2d opencv_perf_ocl opencv_perf_video opencv_perf_highgui opencv_performance opencv_perf_imgproc opencv_perf_photo

So I have decided to write a python script to run these different performace executables and then writing all test results to a csv file.

enienws gravatar imageenienws ( 2013-07-22 02:31:10 -0500 )edit