Shape-Erased Feature Learning for Visible-Infrared Person Re-Identification
This work addresses the modality gap in person re-identification for security and surveillance applications, presenting an incremental improvement by explicitly erasing body shape to force extraction of other shared cues.
The paper tackles the challenge of learning diverse modality-shared semantic concepts for visible-infrared person re-identification by proposing a shape-erased feature learning paradigm that decorrelates features into orthogonal subspaces, achieving enhanced representation diversity as demonstrated through experiments on datasets like SYSU-MM01, RegDB, and HITSZ-VCM.
Due to the modality gap between visible and infrared images with high visual ambiguity, learning \textbf{diverse} modality-shared semantic concepts for visible-infrared person re-identification (VI-ReID) remains a challenging problem. Body shape is one of the significant modality-shared cues for VI-ReID. To dig more diverse modality-shared cues, we expect that erasing body-shape-related semantic concepts in the learned features can force the ReID model to extract more and other modality-shared features for identification. To this end, we propose shape-erased feature learning paradigm that decorrelates modality-shared features in two orthogonal subspaces. Jointly learning shape-related feature in one subspace and shape-erased features in the orthogonal complement achieves a conditional mutual information maximization between shape-erased feature and identity discarding body shape information, thus enhancing the diversity of the learned representation explicitly. Extensive experiments on SYSU-MM01, RegDB, and HITSZ-VCM datasets demonstrate the effectiveness of our method.