---
license: cdla-permissive-2.0
tags:
- table structure recognition
- table extraction
---

# ICDAR-2013.c

The ICDAR-2013.c dataset was released in 2023.

You can think of ICDAR-2013.c as a fork (a modified version, in this case by different authors) of the original ICDAR-2013 dataset from the ICDAR 2013 Table Competition.

It contains:
- manual corrections to minor annotation mistakes in the original dataset
- automated corrections (such as canonicalization) to correct oversegmentation and to make the dataset more consistent with other TSR datasets, like PubTables-1M

For more details about this version (2023) of the dataset and the manual corrections made to the original dataset, please see ["Aligning benchmark datasets for table structure recognition"](https://arxiv.org/abs/2303.00716).

For the code used to create this dataset, see [https://github.com/microsoft/table-transformer](https://github.com/microsoft/table-transformer).

## Citing

If you use this dataset in your published work, please cite:
```
@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}
}
```

## About the original IDCAR-2013 dataset

The original dataset was released as part of the ICDAR 2013 Table Competition.

It can be downloaded [here](https://roundtrippdf.com/en/downloads/) but as of August 2023 accessing the files returns a 403 Forbidden error.

We release a copy of the original dataset but with manual corrections to fix minor annotation mistakes [here](https://huggingface.co/datasets/bsmock/ICDAR-2013-Table-Competition-Corrected).

### Original license

There is no known license for the original dataset, but the data is commonly referred to as "public", and so we interpret this to mean there are no license restrictions on the original data.

According to [this website](https://roundtrippdf.com/en/data-extraction/pdf-table-recognition-dataset/) from Tamir Hassan (as of August 2023): "These documents have been collected systematically from the European Union and US Government websites, and we therefore expect them to have public domain status."

Associated [code](https://github.com/tamirhassan/dataset-tools) for the data for the 2013 competition carries an Apache-2.0 license.