MFEN:Multi-Frequency Expert Network for Visible-Infrared Person Re-ID
Researchers have developed a Multi-Frequency Expert Network (MFEN) to address the challenges in visible-infrared person re-identification (VI-ReID). The network aims to overcome the significant modality discrepancy between visible and infrared images, which is often caused by differing lighting conditions. MFEN utilizes a mixture-of-experts design to adaptively combine information from various frequency bands, enhancing the extraction of identity-relevant details while filtering out lighting variations. The approach is further supported by Random Frequency Augmentation and Frequency Auxiliary Optimization techniques to improve training and robust representation learning. AI
IMPACT Introduces a novel approach to improve person re-identification across different visual spectra, potentially enhancing surveillance and security systems.