Learning Diverse Features with Part-Level Resolution for Person Re-Identification

21 Jan 2020  ·  Ben Xie, Xiaofu Wu, Suofei Zhang, Shiliang Zhao, Ming Li ·

Learning diverse features is key to the success of person re-identification. Various part-based methods have been extensively proposed for learning local representations, which, however, are still inferior to the best-performing methods for person re-identification. This paper proposes to construct a strong lightweight network architecture, termed PLR-OSNet, based on the idea of Part-Level feature Resolution over the Omni-Scale Network (OSNet) for achieving feature diversity. The proposed PLR-OSNet has two branches, one branch for global feature representation and the other branch for local feature representation. The local branch employs a uniform partition strategy for part-level feature resolution but produces only a single identity-prediction loss, which is in sharp contrast to the existing part-based methods. Empirical evidence demonstrates that the proposed PLR-OSNet achieves state-of-the-art performance on popular person Re-ID datasets, including Market1501, DukeMTMC-reID and CUHK03, despite its small model size.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Person Re-Identification CUHK03-C MGN Rank-1 5.44 # 7
mAP 4.20 # 5
mINP 0.46 # 5
Person Re-Identification CUHK03 detected PLR-OSNet MAP 77.2 # 6
Rank-1 80.4 # 6
Person Re-Identification CUHK03 labeled PLR-OSNet MAP 80.5 # 8
Rank-1 84.6 # 8
Person Re-Identification DukeMTMC-reID PLR-OSNet Rank-1 91.6 # 16
mAP 81.2 # 36
Person Re-Identification Market-1501 PLR-OSNet Rank-1 95.6 # 37
mAP 88.9 # 49
Person Re-Identification Market-1501-C PLR-OS Rank-1 37.56 # 3
mAP 14.23 # 3
mINP 0.48 # 3

Methods


No methods listed for this paper. Add relevant methods here