2024-04-11

Last modified: 2024-04-11 21:00:34

Neural net images

After a bit more training, it's now actually getting less interesting: https://www.youtube.com/watch?v=cEa62tBigAU

It looks like it is learning my face at z=0 and Feynman's at z=1, but it is making them much tighter around their z coordinate, so everything is just dark in between. I guess that minimises cross-talk between the jes part of the network and the Feynman part of the network. I was kind of hoping that it would find commonalities between the two pictures and then the fade in between would only be blending the differences.

Maybe it would work better if I put a lot more faces in it.