Person Re-Identification Based on Non-Local and Gated Axial Attention
In order to further improve the performance of person re-identification technology in pra-ctical application,a person re-identification method combining non-local module and gated axial attention module is proposed by introducing self-attention mechanism.The method inserts the self-attention module into the convolution layer of the re-identification network,captures the global and local context information simultaneously through the non-local module,and introduces the axial attention module of the gating mechanism to improve the accuracy of person re-identification.Ablation experiments are carned out on Market 1501 data set to verify the significant improvement of model performance by using both non-local module and gated axial attention module.The experimental results verify the feasibility and advancement of the design,which has certain reference significance for the research in related fields.
Person re-identificationSelf-attentionNon-local attentionGated axial attentionBOT
万方数据
维普
