基于多语义特征互助学习的无监督跨模态行人重识别OA

To address the issue of insufficient handling of feature differences in existing methods for unsupervised cross-modal person re-identification,this paper proposes an unsupervised cross-modal person re-identification method incorporating cluster consistency correction and multi-semantic feature fusion.By removing noisy samples and incorporating common samples,the method ensures that the mixed clusters of samples are more consistent with the characteristics of shallow and deep semantic feature clusters,thereby improving the homogeneity and complete-ness of cluster division.A multi-semantic feature exchange mechanism was introduced to achieve feature alignment and information interaction,enhancing the complementarity between features and reducing modality differences,thus alleviating overfitting.Experimental results demonstrate that the proposed method performs excellently on the SYSU-MM01 and RegDB datasets,significantly improving key metrics and validating its effectiveness and robustness in un-supervised cross-modal person re-identification tasks.