Learning to Learn across Diverse Data Biases in Deep Face Recognition

Chang Liu, Xiang Yu, Yi-Hsuan Tsai, Ramin Moslemi, Masoud Faraki, Manmohan Chandraker, Yun Fu

Keywords: Imbalanced data classification, long-tailed classification, face recognition

Abstract: Convolutional Neural Networks have achieved remarkable success in face recognition, in part due to the abundant availability of data. However, the data used for training CNNs is often imbalanced. Prior works largely focus on the long-tailed nature of face datasets with respect to the number of instances per identity. In this paper, we show that besides the imbalanced class volume distribution, other variations such as ethnicity, head pose, occlusion and blur can also significantly affect accuracy. To address the problem, we propose a sample level weighting approach called Multi-variation Cosine Margin (MvCoM) which orthogonally enhances the conventional cosine loss function to incorporate the importance of training samples. Further, we leverage a learning to learn approach, guided by a held-out meta learning set and use an additive modeling to predict the MvCoM. Extensive experiments on challenging face recognition benchmarks demonstrate the advantages of our method in jointly handling imbalances due to multiple variations.

One-sentence Summary: A novel MvCoM loss design via a meta-learning scheme to jointly target on multiple data bias factors for more balanced face reocgnition.