Li Mingkun ; Xu Peng ; Li Chun-Guang ; Guo Jun

Abstract

In this paper, we address a highly challenging yet critical task:unsupervised long-term person re-identification with clothes change. Existingunsupervised person re-id methods are mainly designed for short-term scenariosand usually rely on RGB cues so that fail to perceive feature patterns that areindependent of the clothes. To crack this bottleneck, we propose asilhouette-driven contrastive learning (SiCL) method, which is designed tolearn cross-clothes invariance by integrating both the RGB cues and thesilhouette information within a contrastive learning framework. To ourknowledge, this is the first tailor-made framework for unsupervised long-termclothes change \reid{}, with superior performance on six benchmark datasets. Weconduct extensive experiments to evaluate our proposed SiCL compared to thestate-of-the-art unsupervised person reid methods across all the representativedatasets. Experimental results demonstrate that our proposed SiCL significantlyoutperforms other unsupervised re-id methods.