Low-Resolution Guided Dual-Branch Fusion Network for Cross-Resolution Person Re-Identification

Cross-resolution person re-identification(CR-ReID) aims to match pedestrian images captured at varying resolutions. Most existing methods simply apply super-resolution (SR) models to enhance image quality, but they fail to distinguish pedestrians from background regions, which limits the recovery of identity-relevant details. To address this issue, we propose a mask mechanism that enhances fine-grained pedestrian detail recovery by adaptively emphasizing informative regions. Additionally, we introduce a Low-Resolution guided Dual-branch (LRD) framework that leverages low-resolution features to reweight global attention, guiding the fused cross-resolution representations to focus on key local areas. Extensive experiments validate the effectiveness and superiority of the proposed method.