Wang Qizao ; Qian Xuelin ; Li Bin ; Chen Lifeng ; Fu Yanwei ; Xue Xiangyang

Abstract

Cloth-changing person re-identification aims at recognizing the same personwith clothing changes across non-overlapping cameras. Advanced methods eitherresort to identity-related auxiliary modalities (e.g., sketches, silhouettes,and keypoints) or clothing labels to mitigate the impact of clothes. However,relying on unpractical and inflexible auxiliary modalities or annotationslimits their real-world applicability. In this paper, we promote cloth-changingperson re-identification by leveraging abundant semantics present withinpedestrian images, without the need for any auxiliaries. Specifically, we firstpropose a unified Semantics Mining and Refinement (SMR) module to extractrobust identity-related content and salient semantics, mitigating interferencefrom clothing appearances effectively. We further propose the Content andSalient Semantics Collaboration (CSSC) framework to collaborate and leveragevarious semantics, facilitating cross-parallel semantic interaction andrefinement. Our proposed method achieves state-of-the-art performance on threecloth-changing benchmarks, demonstrating its superiority over advancedcompetitors. The code is available at https://github.com/QizaoWang/CSSC-CCReID.