Batch Hard Contrastive Loss and Its Application to Cross-view Gait Recognition

Biometric person authentication comprises two tasks: the identification task (i.e., one-to-many matching) and verification one-to-one matching). In this paper, we propose a loss function called batch hard contrastive (BHCn) for deep learning-based task. For purpose, consider mining techniques developed in translate them to More specifically, inspired by triplet losses learn relative distance task, BHCn an absolute that better represents general. Our method preserves identity-agnostic nature of selecting hardest pair samples each identities instead sample. We validate effectiveness proposed cross-view gait recognition using three networks: lightweight input, structure, output network call GEI + CNN (Gait Energy Image Convolutional Neural Network) as well widely used GaitSet GaitGL, which have sophisticated inputs, structures, outputs. trained these networks with publicly available silhouette-based datasets, OU-ISIR Gait Database Multi-View Large Population (OU-MVLP) dataset Institute Automation Chinese Academy Sciences Multiview (CASIA-B) dataset. Experimental results show outperforms other functions, such conventional loss.