Optimization of Vehicle Re-identification Model Based on Vision Transformer
A vehicle re-identification method based on the Vision Transformer framework was proposed to address the challenges of large intra-class variations and high inter-class similarities in vehicle re-identifi-cation tasks.A key region selection module was designed to integrate attention score matrices from Trans-formers,enhancing the focus on discriminative regions of vehicles and reducing the excessive attention weights on local regions.A hybrid loss function was constructed,incorporating contrastive loss and center loss.The introduction of contrastive loss enhanced the model's ability to capture and compare differences between samples,while center loss promoted tighter clustering of samples within the same category,thus improving inter-class sample discrimination.Experimental results validated the effectiveness of the pro-posed method.
万方数据
维普
