LOGOVIT: Local-global vision transformer for object re-identification

Abstract

Object re-identification (ReID) is prone to errors under variations in scale, illumination, complex background, and object occlusion scenarios. To overcome these challenges, attention mechanisms are employed to concentrate on interesting parts of an object to extract better discriminative features. This paper introduces local-global vision transformer (LoGoViT) for object re-identification by learning a hierarchical-level representation from fine-grained (local) to general (global) context features. It comprises two components: (i) shift and shuffle operations generate robust local features, and (ii) local-global module which aggregates the multi-level hierarchy features of an object. Extensive experiments show that our method achieves state-of-the-art on ReID benchmarks. We further investigate effective augmentation operations and discuss how patch modifications can help the model generalize under occlusion. Our code is available at https://github.com/nguyenphan99/LoGoViT