We have build new annotations for the defined Contextualized Table Extraction task, fusing two well known datasets:
- PubLayNet1, a dataset for Document Layout Analysis with 5 different labeld regions;
- PubTables-1M2, a dataset to perform Table Detection, Table Structure Recognition and Functional Analysis.
Tables are really important sources of information for research purposes, and giving them a context (instead of just focusing on them) can help in their extraction. We have been inspired by mainly two works:
- DocBank3, to reformulate the problem as a token-classification task
- AxCell4, to give table a context also for comparable resarch porpuses
You can read more details in our paper: CTE: Contextualized Table Extraction Dataset (under review)
About the PDF data: We do not own the copyright of the original data and we cannot redistribute them. The PDF files can be downloaded from here.
Run in your environment:
pip install -e .
to install dependencies.
After that, download:
- PubLayNet annotations from here
- PubTables-1M-PDF_Annotations_JSON from here And collocate them as described in Project Tree section.
Finally, to generate the annotations, run:
python src/generate_annotations.py
You will find train, val and test annotation json files in the data/merged subfolder.
├── setup.py - Initialization script
├── visualization.ipynb - Visualize annotations on example images
│
├── src/
│ ├── generate_annotations.py - Annotation json files generation
│ └── data/ - folder of scripts used by generate_annotations.py
│
├── data/ - where papers and annotations are stored
│ ├── publaynet/ - train, val, test jsons and PubLayNet_PDF folder
│ ├── pubtables-1m/ - PubTables-1M-PDF_Annotations_JSON folder
│ └── merged/
│ ├── test.json - CTE annotations (as described in config file format section)
│ ├── train.json - CTE annotations (as described in config file format section)
│ └── val.json - (CTE annotations as described in config file format section)
Config files are in .json format. Example:
"objects":
{
"PMC#######_000##.pdf":
[
[0, [157, 241, 807, 738], 1],
[1, [157, 741, 807, 1238], 1],
...
]
...
},
"tokens":
{
"PMC#######_000##.pdf":
[
[0, [179, 241, 344, 271], 'Unfortunately,', 1, 0],
[1, [354, 241, 412, 271], 'these', 1, 0],
[2, [423, 241, 604, 271], 'quality-adjusted', 1, 0],
...
]
...
}
"links":
{
"PMC#######_000##.pdf":
[
...,
[9, 11, [31, 41]],
[10, 12, [22, 23, 24, 25, 26, 27, 28, 29, 30, 31]],
...
]
...
}
Each object has these information:
- object id
- bounding box coordinates
- class id
Each token has these information:
- token id
- bounding box coordinates
- text
- class id
- object id (to which it belongs)
Each link has these information:
- link id
- class id
- token ids (list of tokens linked together)
If you want to use our dataset in your project1, please cite us:
@misc{https://doi.org/10.48550/arxiv.2302.01451,
doi = {10.48550/ARXIV.2302.01451},
url = {https://arxiv.org/abs/2302.01451},
author = {Gemelli, Andrea and Vivoli, Emanuele and Marinai, Simone},
keywords = {Computation and Language (cs.CL), Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {CTE: A Dataset for Contextualized Table Extraction},
publisher = {arXiv},
year = {2023},
copyright = {Creative Commons Attribution Share Alike 4.0 International}
}
Footnotes
-
Xu Zhong et al., PubLayNet: largest dataset ever for document layout analysis, ICDAR 2019. ↩ ↩2
-
B. Smock et al., "Towards a universal dataset and metrics for training and evaluating table extraction models", arXiv, November 2021. ↩
-
Li, Minghao, et al. "DocBank: A benchmark dataset for document layout analysis." arXiv preprint arXiv:2006.01038 (2020). ↩
-
Kardas, Marcin, et al. "Axcell: Automatic extraction of results from machine learning papers." arXiv preprint arXiv:2004.14356 (2020) ↩