Part-Attention Based RGB-Infrared Person Re-Identification Network

Person re-identification is a challenging task in artificial intelligence, particularly in the context of RGB-Infrared images. This paper proposes a method that leverages a backbone network and self-attention mechanism to address the challenges in this domain. The approach achieves significant performance improvement, as demonstrated through experiments on the SYSU-MM01 dataset. The proposed method incorporates a modified ResNet-50 as a feature extractor and utilizes various loss functions to optimize different aspects of the re-identification process. The results show an mAP of 66.37% and high precision scores, highlighting the effectiveness of the approach. This study confirms the potential of correlation-based methods for person re-identification across different domains.