ICDAR-2013.c Dataset

Overview

ICDAR‑2013.c dataset was released in 2023 and can be considered a modified version of the original ICDAR‑2013 dataset. It contains manual corrections of minor annotation errors in the original data as well as automated normalizations to fix over‑segmentation issues and improve consistency with other TSR datasets such as PubTables‑1M.

Content

• Manual Corrections: Small annotation errors in the original dataset are corrected manually.
• Automated Corrections: Normalization is applied to resolve over‑segmentation and increase alignment with other TSR datasets.

Citation

If your research uses this dataset, please cite the following paper:

@article{smock2023aligning,
  title={Aligning benchmark datasets for table structure recognition},
  author={Smock, Brandon and Pesala, Rohith and Abraham, Robin},
  booktitle={International Conference on Document Analysis and Recognition},
  pages={371--386},
  year={2023},
  organization={Springer}
}

Original Dataset

The original ICDAR‑2013 dataset was released for the ICDAR 2013 Table Competition. The original dataset has no known license but is generally considered public domain, so we treat it as having no license restrictions.