QUEST Dataset Repository

This repository contains the DocILE dataset splits and corresponding ground truth annotations used in the paper "QUEST: Quality-aware Semi-supervised Table Extraction for Business Documents". The data is derived from the original DocILE dataset and processed following the methodology described in our paper.

Repository Structure

• DOCILE_QUEST/
  • A-train/
    • GT_train/ # Ground truth annotations for training set
    • img_and_ocr_train/ # Images and OCR files for training set
  • A-val/
    • GT_val/ # Ground truth annotations for validation set
    • img_and_ocr_val/ # Images and OCR files for validation set
  • A-test/
    • GT_test/ # Ground truth annotations for test set
    • img_and_ocr_test/ # Images and OCR files for test set

Dataset Information

The dataset consists of 954 annotated tables split into:

• Training set: 670 tables
• Validation set: 143 tables
• Test set: 141 tables

Each subset directory contains:

• GT_{mode}/: Ground truth annotations in JSON format
• img_and_ocr_{mode}/: Corresponding document images (PNG) and OCR files

Citations

If you use this dataset in your research, please cite:

@misc{thomas2025questqualityawaresemisupervisedtable,
      title={QUEST: Quality-aware Semi-supervised Table Extraction for Business Documents}, 
      author={Eliott Thomas and Mickael Coustaty and Aurelie Joseph and Gaspar Deloin and Elodie Carel and Vincent Poulain D'Andecy and Jean-Marc Ogier},
      year={2025},
      eprint={2506.14568},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2506.14568}, 
}

@inproceedings{vSimsa2023DocILEBF,  
    title={DocILE Benchmark for Document Information Localization and Extraction},  
    author={vStvep'an vSimsa and Milan vSulc and Michal Uvrivc'avr and Yash J. Patel and Ahmed Hamdi and Matvej Koci'an and Maty'avs Skalick'y and Jivr'i Matas and Antoine Doucet and Micka{\"e}l Coustaty and Dimosthenis Karatzas},  
    booktitle={IEEE International Conference on Document Analysis and Recognition},  
    year={2023},  
    url={https://api.semanticscholar.org/CorpusID:256827641}  
}