Person Re-identification Based on Two Dimensional Attention
MAO Feng
LIU Zhigang
CHANG Lele
Abstract：In recent years,person re-identification methods based on attention mechanism mostly apply attention to pedestrian local areas,and the granularity of attention is not fine enough,and the dimension is relatively single.To solve this problem,this pa-per proposes a spatial dimension based attention module,namely two dimensional attention module(TDAM).The module can infer the attention map of the feature map from the spatial dimension,obtain the fine-grained attention parameters of the space in the channel,and then multiply the attention parameters by the input feature map to refine the global adaptive features of pedestrians.TDAM and channel attention(CAM)are combined to form a new attention module--point attention module(PAM),which takes into account the two dimensions of channel and space.The Rank1 of TDAM on Market-1501 and DukeMTMC-ReID data sets reach-es 91.8%and 82.8%respectively,and the map reaches 81.1%and 73.2%respectively.The Rank1 of PAM on Market-1501 and DukeMTMC-ReID data sets reaches 93.3%and 84.1%respectively,and the map reaches 82.6%and 75.0%respectively.The perfor-mance of the algorithm is better than some existing person re-identification methods based on attention mechanism,and even close to some excellent methods.
Keywords：person re-identificationattentionchanneltwo dimensionspace
Publication Date：2025-07-20
Online Publishing Date：2025-09-25(First online date of this platform, not the publication date of the document)
Pages：6( 1897-1902 )