Cross-Modality Person Re-Identification via GAN-based Mixed-Modality Augmentation

Cross-modality Person Re-Identification aims to identify the same individual across color images captured during daytime and infrared images captured at nighttime. Due to the significant visual discrepancy between these modalities, it is considered substantially more challenging than conventional person re-identification. To alleviate the modality gap, extensive studies have been conducted on representation learning, network architecture design, and loss function optimization. In this paper, we focus on data augmentation, which has received relatively limited attention but plays a crucial role in cross-modality learning. Existing augmentation methods primarily aim to increase diversity within a single modality, which limits their ability to adequately reflect the distribution discrepancy across different modalities. To address this limitation, we propose a data augmentation framework that integrates GAN-based modality transformation with mixed-modality information to simultaneously enhance data diversity and effectively reduce modality discrepancies. To validate the effectiveness of the proposed method, we conduct extensive experiments on SYSU-MM01, a widely used benchmark dataset for cross-modality person re-identification, and compare our approach with various state-of-the-art methods. Experimental results demonstrate the superiority of the proposed method.