Multi-scale local-global architecture for person re-identification

Abstract With the emergence of deep learning method, which has been driven a great success for field person re-identification (re-ID). However, existing works mainly focus on first-order attention (i.e., spatial and channels attention) statistics to model valuable information re-ID. On other hand, most methods operate data points respectively, ignores discriminative patterns some extent. In this paper, we present an automated framework named multi-scale local-global The consists two components. first component is that high-order module adopted learn subtle differences among pedestrians generate informative features. novel architecture spectral feature transformation designed make optimization group wise similarities. Furthermore, fuse components together form ensemble Extensive experiments were conducted three benchmark datasets, i.e., Market-1501, DukeMTMC-reID, CUHK03, showing superiority proposed method.