A digital machine-readable dataset of both purely symbolic and OMR-oriented representations of the music pieces in Baishidaoren Gequ.
The dataset KuiSCIMA (Jiang Kui Score Images for Musicological Analysis) is the first machine-readable digital OMR dataset containing ancient Chinese suzipu 俗字谱 notation. It is based on five different influential editions of Jiang Kui's 17 suzipu pieces in the influential collection Baishidaoren Gequ 白石道人歌曲. The JSON format employed in this dataset is semantically close to the original notation and features symbol-level annotations of the textual and musical contents with their positions.
The annotation tool used for viewing, editing and exporting from the KuiSCIMA dataset is available under
https://github.com/SuziAI/gui-tools.
Currently, the version of gui-tools needed is tagged as v2.1.
KuiSCIMA v2.0 is an updated and expanded version of KuiSCIMA v1.0. It contains not only the 17 suzipu pieces,
but all 109 pieces of Baishidaoren Gequ, including lülüpu, and jianzipu musical notations. For all new work, please
cite KuiSCIMA v2.0:
KuiSCIMA v2.0: Improved Baselines, Calibration, and Cross-Notation Generalization for Historical Chinese Music Notations in Jiang Kui's Baishidaoren Gequ (freely available as preprint).
@InProceedings{RepoluskIcdar2025,
author={Repolusk, Tristan and Veas, Eduardo},
editor={Yin, Xu-Cheng and Karatzas, Dimosthenis and Lopresti, Daniel},
title={KuiSCIMA v2.0: Improved Baselines, Calibration, and Cross-Notation Generalization for Historical Chinese Music Notations in Jiang Kui's Baishidaoren Gequ},
booktitle={Document Analysis and Recognition - ICDAR 2025},
year={2025},
publisher={Springer Nature Switzerland},
address={Cham},
pages={116--132},
doi={10.1007/978-3-032-04630-7_7}
}
KuiSCIMA v1.0 should only be cited if you refer to that specific version:
The KuiSCIMA Dataset for Optical Music Recognition of Ancient Chinese Suzipu Notation
@InProceedings{RepoluskIcdar2024,
author={Repolusk, Tristan and Veas, Eduardo},
editor={Barney Smith, Elisa H. and Liwicki, Marcus and Peng, Liangrui},
title={The KuiSCIMA Dataset for Optical Music Recognition of Ancient Chinese Suzipu Notation},
booktitle={Document Analysis and Recognition - ICDAR 2024},
year={2024},
publisher={Springer Nature Switzerland},
address={Cham},
pages={38--54},
doi={10.1007/978-3-031-70552-6_3}
}
For the details of what editions of Baishidaoren Gequ are contained in KuiSCIMA, preprocessing, dataset composition etc., refer to the chapter in my dissertation. See the annotation remarks for additional comments on annotations of unclear notation instances in KuiSCIMA.
See the content list for a complete listing of all the contents in KuiSCIMA.
In addition, the folder artificial_suzipu_dataset contains 36 handwritten (by me) instances of all 77 suzipu classes,
even the ones that do not occur in the original Baishidaoren Gequ. This data is not yet used, since all my attempts to
train models with this artificial data included resulted in worse overall performance.