Graph-Based Soft Fusion of Spatiotemporal Information for Pedestrian Re-identification

This paper addresses the challenges of integrating spatial-temporal information in pedestrian re-identification by introducing a novel graph-based soft fusion method. Our approach effectively reduces the impact of noise introduced by fusing spatial-temporal information by limiting the scope of its influence and utilizing it solely for neighbor selection. This method leverages the inherent graph structure of pedestrian re-identification datasets and provides a robust solution for improving the accuracy of person re-identification in challenging scenarios.