Fig. 1

The dataflow of a deep face recognition system. During training the network learns feature representations (f_i) of faces by getting gradually penalized by the loss function. During testing, the pre-trained model is used to generate features of test faces. The generated features are classified/compared for identity determination