As far as I know,it's not available.But it's available on openCV.
yes, I have seen it and it is here:
author says at the bottom of the block that it is designed mainly for nVidia. And that images are not supported in open cl (what is obviously not true). I haven't seen/touch the code yet, but it looks like it's not fully trained as open cv solution.
For AMD developer summit there are few articles of face detection development:
http://openclnews.com/blog/category/coding (search for face recognition there), as well as interview here:
what make it looks like there is some activity. However, if you go to viiewdle' web site www.viewdle.com, what suppose to specialize on it you may find almost nothing there related to subject.
well maybe he didn't check if AMD CPU support image now. on other hand i downloaded that code and successfully compiled it. (with little change in OpenCV include) it worked on my E-350 netbook. it runs on 7-8 fps. execution of GPU parts was only 20-50 ms so it is CPU bound on my netbook. oh and video resolution was 640x480
I wrote one in socles, it doesn't implement the full opencv haar cascade options (the branching within a test or rotated regions) but it has the basic frontal-face stuff.
It's probably the simplest implementation you'll find if you want to understand how haar cascades actually work (it's decidedly simpler when you remove the xml snot and mathematician-authored-code), and socles also has a pretty good (imho) integral image calculator. opencloovision, faint, and opencv helped me work out some of the fiddly bits.
The cascade doesn't run particularly fast on cayman (and believe me, i tried), but was fairly decent on an nvidia GTX 480 (<10ms total for vga, depending on scanning step scale/size), when i get a 7970 card soon I will revisit it and i'm hoping GCN will help it work better.
socles is GPLv3 and uses JOCL (i.e. java).