How do you compute abs() using UMat?

epatton
19 ●1 ●5

Hi My input is UMat 32FC1 and I'd like to compute abs() but I get a compile error "no matching function for call to 'abs'". my code works find for Mat, but I'd like to use the GPUs. I also tried convertScaleAbs() but it outputs an 8 bit value whereas I want to stay with 32FC1. Any ideas on what's the most efficient way to do this other than writing my own ocl kernel? Many Thanks

add a comment

1 answer

Sort by » oldest newest most voted

answered Sep 26 '17

matman
1001 ●6 ●16

Mmh, cv::abs doesn't seem to exist anymore. Even in docs the link is dead. As a workaround you could try:

cv::absdiff(src, cv::Scalar::all(0), dst);

Comments

@matman I think it is abs and it is not UMat compatible

LBerger (Sep 26 '17)edit

Thanks, it works. Odd thing is that absdiff is marginally faster than copying the UMat to mat and then using abs(). Odd in that doing so doesn't use the GPU yet it runs almost as fast.

aUmat.copyTo(aMat); bMat = cv::abs(aMat); bMat.copyTo(bUMat);