Captured images generated in vvvv, and added some text.

So I found this library: https://github.com/ShiqiYu/libfacedetection
And it was super easy to use. So after a wrapper class and a handling node here are the results from vvvv with any DX11 texture:

Trololo at 60-70 fps with multiview-reinforce engine.

Crowd footage at 10-20 fps with multiview engine. It tracks as many faces as much you throw at it.

This library is fast but far from accurate. But I made the wrapper class in a way that in the future it can be easily extended with other engines, so if anyone finds robust and fast face trackers, throw that at me and I'll implement it if they don't cost a fortune.

Available in mp.dx http://vvvvpm.github.io/#mp.dx
Enjoy!

SkeletalMesh (DX11 Assimp) and SkeletonSimple (Assimp) in mp.dx are now using the Skeleton interfaces V2 implementation. This means you can have spreadable skeletal rigged animations without the need for FBX4V. matka was nagging me with it for a while now and finally managed to find some time for it. Get mp.dx with vpm from https://vvvvpm.github.io/#mp.dx and overwrite all packs (especially mp.essentials and mp.fxh)