Oracle bone inscription detection

The goal of this paper is to serve as a guide for selecting a detection architecture for Oracle Bone Inscription (OBI) that achieves the right speed/memory/accuracy balance for a given platform. Many successful systems have been proposed in recent years, but directly migrating these methods to OBI data may lead to unsatisfying performance due to the corrosion, noise, and distribution. We present a unified implementation of the Faster R-CNN 1, SSD 2, YOLOv3 3, RFBnet 4, and RefineDet 5 systems which we view as experiment architectures and trace out the speed/accuracy trade-off. The method with the best overall performance was selected as the baseline. Then a series of improvements were made according to the characteristics of the data. The experiment shows that our method reaches the F-measure of 80.1, nearly 2% higher than the baseline. Experimental data and algorithms will soon be available at http://jgw.aynu.edu.cn.