flim

I'm a bit confused: whilst the help file shows performance "humps" as soon as one works with [jit.pwindow]s above 160x120 (respectively ignoring resize), these 2 patches seem to be a lot busier in the bg - without sacrifice in performance..

cv-jit-faces-performance

Those two patches are somewhat complicated, but if you look carefully, you'll see that nesa sends cv.jit.faces very small matrices (80 x 60) and in my patch, although I send larger matrices to cv.jit.faces, most of the time, it's not doing anything!

I don't know if you mis-typed, but jit.pwindow has nothing to do with performance in this situation. cv.jit.faces is an expensive object, processor-wise, so you either need a small input, as in nesa's patch, or to only send large matrices once in a while, when you really need it.

I only thought that jit.pwindow's a good way to tell how task-intensive cv.jit objects are..

you're eye recognition patch might take me a while, so I'll take the small input "shortcut" by nesa.

I should be fine by basically giving the helpfile some [jit.matrix] resizing on the LCD side & rejoining my detected region and my webcam matrix with [jit.op], right?

ok, this is the furthest I'll get today. after 7h of comprehension I'm done :)

ok, I give up. spent all day downgrading jean-marc's eye recognition patch as well as nesa's fine work, thereby trying to understand which parts actually reduce performance humps (enabling a fluent picture). but they never work after I've done that, although they should (going by my personal grasp of logic). either some bug's trying to torture me or I'm just nuts. I'm only trying to give the help file the same "performance" as in the patches above.

Hi flim where can I find this jean-marc's eye recognition patch?

Hi flim where can I find this jean-marc's eye recognition patch?
I`m interested to try it.

cv.jit.faces: performance