Face Feature Embedding

A facial feature typically represents the original image in an embedding space, where the distance between embeddings is used to measure face similarity. To implement face verification and recognition at scale, it is indispensable to have a discriminative face embedding in which faces of the same person have small distances and faces of different people have large distances. Once such embedding is obtained, face verification involves thresholding the distance between the two embeddings, and face recognition becomes a k-Nearest Neighbor (k-NN) classification problem that takes feature embeddings as inputs rather than original images.