MatchTemplate OCL and CPU Timing

asked 2015-12-09 11:58:16 -0600

Honest_Trefor gravatar image

Hi,

I was looking for a performance boost using the OCL implementation of matchtemplate(), but the Mat and UMat implementation both take the same time. When diving a little deeper, It is the normalization that takes the majority of the time, anyone have anymore information on how to implement the fastest normalized template matching?

Results:

CPU
TM_CORR: 16ms
TM_CORR_NORMED: 47ms

OCL
TM_CORR: 0ms
TM_CORR_NORMED: 47ms

edit retag flag offensive close merge delete

Comments

After diving a little deeper, I am not convinced the opencl normalized functions are implemented for a 32bit float.

Honest_Trefor gravatar imageHonest_Trefor ( 2015-12-10 13:38:17 -0600 )edit
add a comment