Clustering Refinement Based on Camera-View for Unsupervised Person Re-Identification
LING Zixuan
JIN Zhong
Abstract：Most existing unsupervised person re-identification methods follow a clustering-based strategy,which alternates between generating pseudo labels by a clustering algorithm and training a person Re-ID model based on these pseudo labels.Howev-er,for the procedure of clustering,they ignore the feature variances of image under the change of camera-views.That is,samples within one identity may be gathered into multiple clusters according to their camera labels.Therefore,this paper proposes to rectify the instance pairs'similarity with camera-views,making them more closer to ones without the effect of the change of camera-views,which greatly facilitates the the following clustering and model training.Experiments on existing benchmark datasets demonstrate that the method is superior to most unsupervised counterparts.
Keywords：person Re-IDclusteringpseudo labelcamera view
Publication Date：2025-02-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：5( 384-388 )