Contrastive Bayesian Analysis for Deep Metric Learning

Kan，Shichao1; He，Zhiquan2; Cen，Yigang3; Li，Yang4; Mladenovic，Vladimir5; He，Zhihai6

2022

10.1109/TPAMI.2022.3221486

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE   Impact Factor & Quartile

0162-8828

1939-3539

Recent methods for deep metric learning have been focusing on designing different contrastive loss functions between positive and negative pairs of samples so that the learned feature embedding is able to pull positive samples of the same class closer and push negative samples from different classes away from each other. In this work, we recognize that there is a significant semantic gap between features at the intermediate feature layer and class labels at the final output layer. To bridge this gap, we develop a contrastive Bayesian analysis to characterize and model the posterior probabilities of image labels conditioned by their features similarity in a contrastive learning setting. This contrastive Bayesian analysis leads to a new loss function for deep metric learning. To improve the generalization capability of the proposed method onto new classes, we further extend the contrastive Bayesian loss with a metric variance constraint. Our experimental results and ablation studies demonstrate that the proposed contrastive Bayesian metric learning method significantly improves the performance of deep metric learning in both supervised and pseudo-supervised scenarios, outperforming existing methods by a large margin.

Analytical models Bayes methods Bayesian analysis deep metric learning Information science Measurement representation learning Semantics similarity and distance learning Task analysis Training

EI

English

Others

20224613123839

Deep learning ; Distance education ; Job analysis ; Personnel training

Ergonomics and Human Factors Engineering:461.4 ; Personnel:912.4

ENGINEERING

2-s2.0-85141550015

Scopus

1.School of Computer Science and Engineering, Central South University, Changsha, Hunan, China
2.Guangdong Multimedia Information Service Engineering Technology Research Center, Shenzhen University, China
3.Institute of Information Science, School of Computer and Information Technology, Beijing Jiaotong University, Beijing, China
4.Department of Electrical Engineering and Computer Science, University of Missouri, Columbia, MO, USA
5.Faculty of Technical Sciences University of Kragujevac, Cacak, Serbia
6.Southern University of Science and Technology, Shenzhen, China

Kan，Shichao,He，Zhiquan,Cen，Yigang,et al. Contrastive Bayesian Analysis for Deep Metric Learning[J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,2022,PP(99):1-18.

Kan，Shichao,He，Zhiquan,Cen，Yigang,Li，Yang,Mladenovic，Vladimir,&He，Zhihai.(2022).Contrastive Bayesian Analysis for Deep Metric Learning.IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,PP(99),1-18.

Kan，Shichao,et al."Contrastive Bayesian Analysis for Deep Metric Learning".IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE PP.99(2022):1-18.