Scholarlink

<p>Person reidentification (ReID) is crucial in intelligent surveillance and public security; yet, it faces challenges in cloth-changing scenarios. Existing methods struggle with distinguishing identities under varying clothing and insufficient fine-grained feature utilization. To address these limitations, we propose a dynamic fine-grained optimization via semantic-driven augmentation framework for cloth-changing person ReID. This framework includes a semantic-driven clothing diversity augmentation module using a diffusion model to generate diverse, identity-consistent clothing images, and a dynamic feature recomposition module to enhance adaptability to clothing changes. Additionally, a granularity-aware loss function reduces clothing dependency by minimizing mutual information between local features. Experimental results on benchmark datasets demonstrate our method’s superior performance and robustness in complex cloth-changing scenarios compared to existing methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dynamic fine-grained optimization via semantic-driven augmentation for robust cloth-changing person reidentification

  • 期刊论文

  • 发表日期：

    2026年1月11日

查看全文
  • Mengnan Hu,
  • Wenjing Zhang,
  • Rong Wang

摘要

Person reidentification (ReID) is crucial in intelligent surveillance and public security; yet, it faces challenges in cloth-changing scenarios. Existing methods struggle with distinguishing identities under varying clothing and insufficient fine-grained feature utilization. To address these limitations, we propose a dynamic fine-grained optimization via semantic-driven augmentation framework for cloth-changing person ReID. This framework includes a semantic-driven clothing diversity augmentation module using a diffusion model to generate diverse, identity-consistent clothing images, and a dynamic feature recomposition module to enhance adaptability to clothing changes. Additionally, a granularity-aware loss function reduces clothing dependency by minimizing mutual information between local features. Experimental results on benchmark datasets demonstrate our method’s superior performance and robustness in complex cloth-changing scenarios compared to existing methods.