ABSTRACT
In this chapter, we discuss the object re-identification problem in the computer vision context of vehicles, people, etc. This task allows us to match objects under different views, which benefits many real-world applications such as public safety and traffic control. Existing works in this area focus on the development of discriminative features, ranging from backbone to loss function design. We also highlight the necessity of data-centric research, where improving and analysing data is the primary objective. Apart from these technical perspectives, we discuss responsible practices, such as the use of synthetic data instead of real data, data anonymisation and economical learning schemes. These practices would support ethical use and deployment of object re-identification systems.
