Interact, Embed, and EnlargE: Boosting Modality-Specific Representations for Multi-Modal Person Re-identification

Multi-modal person Re-ID introduces more complementary information to assist the traditional task. Existing multi-modal methods ignore importance of modality-specific in feature fusion stage. To this end, we propose a novel method boost representations for Re-ID: Interact, Embed, and EnlargE (IEEE). First, cross-modal interacting module exchange useful between different modalities extraction phase. Second, relation-based embedding enhance richness descriptors by global into fine-grained local information. Finally, margin loss force network learn each modality enlarging intra-class discrepancy. Superior performance on dataset RGBNT201 three constructed datasets validate effectiveness proposed compared with state-of-the-art approaches.