Some text-type commentary over some work in progress to explain a coding experiment in which I use facial recognition to measure the appearance of faces in random noisy meshes, to propel a genetic algorithm into breeding more face-like meshes. The 'genes' are the mesh vertex coordinates - strong genes get passed on to future generations, so that the population slowly evolves into heads scoring high on facial recognition - supposedly looking more and more like a recognisable face to humans too. The faces however can be disappointingly weak to human observation. I'm moving the project in another direction so this video is here to quickly document my attempts.
I have to credit Phil McCarthy for achieving something similar in 2012 using a hill climbing algorithm and 2D drawing. boingboing.net/2012/08/05/pareidoloop.html
Loading more stuff…
Hmm…it looks like things are taking a while to load. Try again?