Multi-Modal Object Re-Identification with Dual Semantic Guidance and Global-Local Mutual Modulation

Researchers proposed a new multi-modal object re-identification framework that injects text semantics and modulates global-local features.

The method is designed to reduce background clutter and cross-modal misalignment while distinguishing target instances more reliably.

Its key components are the Text-Semantic Injector, Masked Global-Local Modulator, and Hierarchical MoE Fusion.

The paper reports improved results on three benchmarks, suggesting stronger multi-modal matching and retrieval performance.