This example is based on VGGFace model to get a person face features and store as pre-trained features.
Real-time image then compared with every pre-computed label features. I suspect, it won't work
fast on thousands faces due to O(n) complexity, where n
is a number of labels.
Arcticle descring the approach to identify faces: Face Identification with VGGFace and OpenCV
See modules here:
See main two scripts for details: