Comprehensive Detail Refinement Network for Vehicle Re-identification

Abstract

A novel comprehensive detail refinement network, called the CDRNet, to learn robust and diverse features from vehicle images is proposed. There are three modules in the proposed algorithm: the global attention, the detail, and the local feature refinement modules. The global attention module extracts crucial global characteristics while the detail and local refinement modules retrieve important minor features. Experiments on benchmark datasets, VeRi-776 and VehicleID, show that the proposed network outperforms state-of-the-art approaches and is very helpful for vehicle re-identification.