Exploiting robust unsupervised video person re‐identification

Unsupervised video person re-identification (reID) methods usually depend on global-level features. Many supervised reID employed local-level features and achieved significant performance improvements. However, applying to unsupervised may introduce an unstable performance. To improve the stability for reID, this paper introduces a general scheme fusing part models learning. In scheme, feature is divided into equal feature. A local-aware module explore potentials of global-aware proposed overcome disadvantages Features from these two modules are fused form robust representation each input image. This has advantages without suffering its disadvantages. Comprehensive experiments conducted three benchmarks, including PRID2011, iLIDS-VID, DukeMTMC-VideoReID, results demonstrate that approach achieves state-of-the-art Extensive ablation studies effectiveness robustness module. The code generated available at https://github.com/deropty/uPMnet.