An Introduction to Domain Adaptive Object Detection from Synthesis to Reality

In recent years, deep learning based object detection has achieved great progress. These methods typically assume that large amount of labeled training data is available, meanwhile, training and test data are independent and identically distributed. However, the two assumptions are not always hold in practice. In many applications, it is prohibitively expensive and time-consuming to obtain large quantities of labeled data. Using computer graphics technology to generate a large number of labeled data provides a solution to this problem. Unfortunately, direct transfer across domains from synthesis to reality often performs poorly due to the presence of domain shift. Domain adaptive object detection are concerned with accounting for these types of challenges. In this paper, we present an introduction to these fields. Firstly, we briefly introduce the object detection and domain adaptation. Secondly, the synthetic object detection datasets and related software tools are summarized. Thirdly, we present a categorization of approaches, divided into discrepancy-based methods, adversarial discriminative methods, reconstruction-based methods and others. Finally, we also discuss some potential deficiencies of current methods and several open problems which can be explored in future work.