Multi-Granularity Attribute Prompt Learning for Cloth-Changing Person Re-Identification
作者
X W Li,Hefei Ling,Shijuan Huang,Chenxing Zhao
出处
期刊: [Institute of Electrical and Electronics Engineers] 日期:2026-04-21卷期号:: 10942-10946
标识
DOI:10.1109/icassp55912.2026.11464029
摘要
Cloth-changing person re-identification (CC-ReID) addresses the practical need to recognize pedestrians across cameras after clothing changes. Existing methods use biological traits and clothing labels to extract invariant features but under-utilize semantic attributes and require costly manual annotation. To overcome these limitations, we propose a Multi-granularity Attribute Prompt Learning (MAPL) method. MAPL incorporates visual deep prompts in a dual-branch network (raw and clothing-erased images) to learn clothing-irrelevant representations. It generates pseudo clothing labels through attribute recognition to reduce annotation dependency. Using these pseudo-labels and attribute texts, it enhances identity-related attributes while suppressing clothing-related ones. This is achieved through feature disentanglement at both the caption and attribute levels. Experiments show that MAPL achieves state-of-the-art performance, outperforming the second-best method by 3.2%/13.0% in Rank-1 and 4.0%/18.3% in mAP on PRCC and LTCC under clothing-changing settings. The source code is released at https://github.com/IPMD2025/MAPL.