Scholarlink

<p>Visible-infrared person reidentification (VI-ReID) aims to match person images across different modalities. Traditional VI-ReID methods focus on intra-modality feature relationships, ignoring inter-modality correlations. To address this, we propose a modality-aware graph convolutional network (MAGCN) that simultaneously mines intra-modality and inter-modality relationships. MAGCN comprises three key modules: a modality-aware module (MAM) to extract modality-specific features, an intra-modality relationship graph module (IRGM) to enhance feature representation within modalities, and a cross-modality feature alignment module (CFAM) to achieve cross-modality matching. Extensive experiments on two public datasets demonstrate that MAGCN outperforms state-of-the-art methods, achieving higher accuracy in cross-modal person reidentification. The code and datasets are available at: <a href="https://github.com/CQNU-ZhangLab/SYSU">https://github.com/CQNU-ZhangLab/SYSU</a>.</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

Modality-aware graph CNN for cross-modal person reidentification

  • 期刊论文

  • 发表日期：

    2025年3月22日

查看全文
  • Ruisheng Ran,
  • Yu He,
  • Wei Hu,
  • Wenfeng Zhang,
  • Qibing Qin,
  • Shuang Chen

摘要

Visible-infrared person reidentification (VI-ReID) aims to match person images across different modalities. Traditional VI-ReID methods focus on intra-modality feature relationships, ignoring inter-modality correlations. To address this, we propose a modality-aware graph convolutional network (MAGCN) that simultaneously mines intra-modality and inter-modality relationships. MAGCN comprises three key modules: a modality-aware module (MAM) to extract modality-specific features, an intra-modality relationship graph module (IRGM) to enhance feature representation within modalities, and a cross-modality feature alignment module (CFAM) to achieve cross-modality matching. Extensive experiments on two public datasets demonstrate that MAGCN outperforms state-of-the-art methods, achieving higher accuracy in cross-modal person reidentification. The code and datasets are available at: https://github.com/CQNU-ZhangLab/SYSU.