Cloth-Changing Person Re-identification from A Single Image with Gait Prediction and Regularization

Cloth-Changing person re-identification (CC-ReID) aims at matching the sameperson across different locations over a long-duration, e.g., over days, andtherefore inevitably meets challenge of changing clothing. In this paper, wefocus on handling well the CC-ReID problem under a more challenging setting,i.e., just from a single image, which enables high-efficiency and latency-freepedestrian identify for real-time surveillance applications. Specifically, weintroduce Gait recognition as an auxiliary task to drive the Image ReID modelto learn cloth-agnostic representations by leveraging personal unique andcloth-independent gait information, we name this framework as GI-ReID. GI-ReIDadopts a two-stream architecture that consists of a image ReID-Stream and anauxiliary gait recognition stream (Gait-Stream). The Gait-Stream, that isdiscarded in the inference for high computational efficiency, acts as aregulator to encourage the ReID-Stream to capture cloth-invariant biometricmotion features during the training. To get temporal continuous motion cuesfrom a single image, we design a Gait Sequence Prediction (GSP) module forGait-Stream to enrich gait information. Finally, a high-level semanticsconsistency over two streams is enforced for effective knowledgeregularization. Experiments on multiple image-based Cloth-Changing ReIDbenchmarks, e.g., LTCC, PRCC, Real28, and VC-Clothes, demonstrate that GI-ReIDperforms favorably against the state-of-the-arts. Codes are available athttps://github.com/jinx-USTC/GI-ReID.