Three Stages, One Flow

A face recognition system takes an input photograph and produces an identity label through three stages. Detection locates the face and crops it into a bounding box. The embedding stage converts that cropped face into a fixed-length numeric vector. The matching stage compares that vector against stored vectors and returns the closest identity. Each stage hands a specific, narrower piece of data to the next: the whole photo becomes a face crop, the face crop becomes a vector, and the vector becomes a name or a no-match result.