Congratulations to Ph.D. Student Qixian Zhang’s Paper “Reliable Pseudo-supervision for Unsupervised Domain Adaptive Person Search” Accepted by TIP
Recently, the paper “Reliable Pseudo-supervision for Unsupervised Domain Adaptive Person Search” by Ph.D. student Qixian Zhang, under the guidance of Prof. Duoqian Miao, has been accepted by IEEE Transactions on Image Processing (TIP), one of the top journals in artificial intelligence. This study targets key challenges in unsupervised domain adaptation person search, including spectral discrepancies across domains and the accumulation of noise in pseudo-label supervision, and proposes a reliable pseudo-supervision enhancement framework (RPPS). The proposed method achieves superior performance on benchmarks such as CUHK-SYSU and PRW, and significantly improves cross-domain robustness.
近日，实验室苗夺谦教授指导的2021级博士研究生张齐贤的文章“Reliable Pseudo-supervision for Unsupervised Domain Adaptive Person Search”被国际顶级期刊《IEEE Transactions on Image Processing》（TIP）正式录用。TIP是图像处理领域公认的顶级期刊，长期位列中科院一区Top期刊，同时为CCF A类与CAAI A类推荐期刊，在国际学术界具有重要影响力。
该研究围绕跨域无监督行人搜索场景下所面临的频谱差异与伪标签噪声累积等关键问题，构建了一种可靠伪监督增强框架（RPPS）。在特征层引入双分支小波增强机制，显式分解并差异化增强低/高频成分，以提升跨域稳定性；在代理层提出动态置信度加权聚类代理更新策略，通过在线—离线两阶段更新抑制噪声代理并稳定聚类中心。所提方法在 CUHK-SYSU 与 PRW 等基准上取得更优性能，并显著提升了跨域鲁棒性。
