Long-Term Cloth-Changing Person Re-identification

Person re-identification (Re-ID) aims to match a target person across camera views at different locations and times. Existing Re-ID studies focus on the short-term cloth-consistent setting, under which re-appears in with same outfit. A discriminative feature representation learned by existing deep models is thus dominated visual appearance of clothing. In this work, we much more difficult yet practical setting where matching conducted over long-duration, e.g., days months therefore inevitably new challenge changing clothes. This problem, termed Long-Term Cloth-Changing (LTCC) understudied due lack large scale datasets. The first contribution work LTCC dataset containing people captured long period time frequent clothing changes. As second contribution, propose novel method specifically designed address cloth-changing challenge. Specifically, consider that cloth-changes, soft-biometrics such as body shape would be reliable. We, therefore, introduce embedding module well cloth-elimination shape-distillation aiming eliminate now unreliable features information. Extensive experiments show superior performance achieved proposed model dataset. code will available https://naiq.github.io/LTCC_Perosn_ReID.html.