Comparative Analysis of Different Loss Functions for Deep Face Recognition

Abstract

Face Recognition has been one the fastest emerging field in the last ten years. Convolutional neural network (CNN) or Deep convolutional neural network (DCNN) have significantly developed the extraordinary state-of-the-art solution for Face Recognition. This promising development results from the enhanced learning and representation of the discriminative features. The learning depends extensively on the loss function employed in the model. The loss function plays a vital role in the training of CNN and its job is to evaluate the performance of the model, i.e. bad performance results in a huge loss and vice versa. The gradients of this loss function are further used in the back propagation of errors which in turn enables the model to improve its learning from the given data. The objective of this paper is to have a comparative analysis of different loss functions available for the Deep Face Recognition.