|About this Abstract
|7th World Congress on Integrated Computational Materials Engineering (ICME 2023)
|Discriminative Object Tracking by Domain Contrast
|Huayue Cai, Xiang Zhang, Long Lan, Changcheng Xiao, Chuanfu Xu, Jie Liu, Zhigang Luo
|On-Site Speaker (Planned)
Multi-domain tracking method improves object tracking by sharing domain information whilst learning private information. In that context, each video sequence as a specific domain serves for a domain-specific layer. We observe an finding that target features from different domains are highly confused with each other. To this end, we propose a domain interaction training paradigm called domain contrast to boost discriminative features by effectively using amounts of instances from all the domains in two novel aspects: 1) a memory-saving training algorithm is proposed to solve the "out-of-the-memory" problem, and 2) a composite class-balanced loss is explored to tackle a imbalanced problem, which not only involves the usual class imbalance problem but also accounts for the case of the totally mere negative instances. Experiments on multiple tracking benchmarks show that our mechanism consistently achieves the tracking performance gain of both base multi-domain tracker and its real-time variant.