Abstract
In recent years, the angle-based softmax losses have significantly improved the performance of face recognition whereas these loss functions are all based on cosine logit. A potential weakness is that the nonlinearity of the cosine function may undesirably saturate the angular optimization between the features and the corresponding weight vectors, thereby preventing the network from fully learning to maximize the angular discriminability of features. As a result, the generalization of learned features may be compromised. To tackle this issue, we propose a Linear-Cosine Softmax Loss (LinCos-Softmax) to more effectively learn angle-discriminative facial features. The main characteristic of the loss function we propose is the use of an approximated linear logit. Compared with the conventional cosine logit, it has a stronger linear relationship with the angle on enhancing angular discrimination through Taylor expansion. We also propose an automatic scale parameter selection scheme, which can conveniently provide an appropriate scale for different logits without the need for exhaustive parameter search to improve performance. In addition, we propose a margin-enhanced Linear-Cosine Softmax Loss (m-LinCos-Softmax) to further enlarge inter-class distances and reduce intra-class variations. Experimental results on several face recognition benchmarks (LFW, AgeDB-30, CFP-FP, MegaFace Challenge 1) demonstrate the effectiveness of the proposed method and its superiority to existing angular softmax loss variants.
Highlights
In recent year, due to the advances of deep convolutional neural networks (CNNs) [1]–[3], the availability of largescale face training data [4], [5] and sophisticated loss function designs [13], [20]–[22], [30], face recognition has achieved significant progress
Since many of today’s deep learning-based face recognition models can achieve beyond 98% verification accuracy on LFW, we used more challenging verification benchmarks, CFP-FP and AgeDB30, for a better performance evaluation
The CFP dataset is for pose-invariant face verification with 7000 images of 500 identities, while the AgeDB dataset is for age-invariant face recognition with 16488 images of 568 identities
Summary
Due to the advances of deep convolutional neural networks (CNNs) [1]–[3], the availability of largescale face training data [4], [5] and sophisticated loss function designs [13], [20]–[22], [30], face recognition has achieved significant progress. Face identification aims to recognize the identity of a target face from a set of registered faces, while face verification aims to verify whether two faces belong to the same identity. It is critical that the trained CNN network must be able to extract discriminative facial features to achieve outstanding recognition performance. Existing loss functions for deep representation learning roughly fall into two categories: metric learningbased methods and classification-based methods
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call