Embed Notice
HTML Code
Corresponding Notice
- Embed this notice
iced depresso (icedquinn@blob.cat)'s status on Monday, 15-Apr-2024 04:52:01 JSTiced depresso @sun @strypey these things typically use convolution networks to guess at some like 40 facial features, and if you are lucky it goes in to some autoencoder to make a fingerprint of those features, which on a good day will just tell you the face is most similar to something else in the database.
if you ever played with vtube software you can see how they're really jittery too