HUST-OBS Oracle Bone Recognition Dataset

The HUST-OBS dataset was constructed in 2024 by researchers from Huazhong University of Science and Technology, South China University of Technology, University of Adelaide, and Anyang Normal University.An open dataset for oracle bone script recognition and decipherment"

The dataset contains 77,064 images of 1,588 decrypted scripts and 62,989 images of 9,411 undecrypted characters, totaling 140,053 images. These images come from three different sources, including books, websites, and existing databases. It is one of the largest OBS identification and decryption datasets to date. In addition, all images and labels have been reviewed and corrected by oracle bone script research experts. The research team hopes that this dataset will inspire and help future research to decipher unknown OBS (oracle bone script).