Learning Instance-Adaptive Low-Rank Orthogonal Subspaces for Clothes-Changing Person Re-Identification
Two new research papers tackle the challenge of clothes-changing person re-identification (CC-ReID), a problem where individuals need to be recognized despite variations in their attire. The first paper, Ortho-ReID, introduces a transformer-based method to model a low-rank clothing subspace and extract clothing-invariant representations through geometric constraints. The second paper, Causal Clothes-Invariant Learning (CCIL), proposes a causality-based approach to prevent models from learning spurious correlations between clothing and identity, thereby improving generalization to unseen clothing. AI
IMPACT These papers advance techniques for person re-identification by addressing the challenge of clothing variations, potentially improving surveillance and security systems.