The face montage building trains a linear classifier on top of the pretrained embedding network - it’s the portion that talks about creating a training folder o

The face montage building trains a linear classifier on top of the pretrained embedding network - it’s the portion that talks about creating a training folder of your face. But yes the photomosiacs strictly don’t use ML, unless you count the internal fun stuff Faiss (the similarity search lib) does to construct fast indexes.