ISFNet: Enhancing Cross-Modal Person Re-Identification via Intermediate Shared Feature Learning
Cross-modal person re-identification between visible and infrared domains remains a challenging problem due to significant modality gaps. This paper presents a novel approach termed Intermediate Shared Feature Network (ISFNet) that explicitly addresses this issue by exploiting intermediate feature representations within a dual-stream backbone. Unlike conventional methods that primarily focus on final-layer features, ISFNet introduces two complementary components: a Multi-layer Feature Cascade Module (MFCM) that aggregates discriminative features across different network stages, and a Dual Feature Generation Module (DFGM) that creates diverse intermediate representations through Instance-Batch Normalization variants. By integrating these modules, ISFNet effectively bridges the cross-modal gap and improves matching accuracy. Comprehensive experiments on the SYSU-MM01 and RegDB datasets demonstrate that the proposed method achieves competitive performance against state-of-the-art approaches, with noticeable improvements in both Rank-1 accuracy and mean average precision.