MixER: From Cross-Modal to Mixed-Modal Visible-Infrared Re-Identification

Visible-infrared person re-identification (VI-ReID) aims to match individuals across different camera modalities, a critical task in modern surveillance. While most existing methods focus on cross-modality matching, real-world systems often involve mixed galleries containing both V and I images, where state-of-the-art methods struggle due to large domain shifts and low discrimination across modalities. These challenges arise because same-modality gallery images may have smaller domain gaps but correspond to different identities. To address this, we propose more comprehensive and challenging mixed-modal evaluation settings that better reflect real-world conditions. This paper also introduces the Mixed Modality-Erased and -Related (MixER) method, which disentangles modality-specific and modality-shared identity information through orthogonal decomposition, modality con-fusion, and ID-modality-related objectives. MixER improves feature robustness across modalities, improving performance in both cross- and mixed-modal settings. Extensive experiments on SYSU-MM01, RegDB, and LLCM show that MixER can achieve state-of-the-art performance with a single backbone and displays strong versatility across diverse mixed-modal scenarios. Our code is available: https://github.com/alehdaghi/MixVI-ReID.