Jiuxiang Gu    Xiangxi Shi    Jason Kuen    Lu Qi    Ruiyi Zhang    Anqi Liu    Ani Nenkova    Tong Sun

ICLR 2024 (Poster)

Research in document image understanding is hindered by limited high-quality document data. ADOPD introduces a large-scale dataset for document page decomposition with a data-driven document taxonomy discovery process during collection, and dense annotations. It supports four tasks: Doc2Mask, Doc2Box, Doc2Tag, and Doc2Seq.

• Data-driven taxonomy discovery during collection to improve diversity and balance.
• Dense annotations including entity masks and text bounding boxes; plus tags/captions cleaned with human involvement.
• Four benchmark tasks: Doc2Mask / Doc2Box / Doc2Tag / Doc2Seq.

---

BibTeX

@inproceedings{
  gu2024adopd,
  title={{AD}o{PD}: A Large-Scale Document Page Decomposition Dataset},
  author={Jiuxiang Gu and Xiangxi Shi and Jason Kuen and Lu Qi and Ruiyi Zhang and Anqi Liu and Ani Nenkova and Tong Sun},
  booktitle={The Twelfth International Conference on Learning Representations},
  year={2024},
  url={https://openreview.net/forum?id=x1ptaXpOYa}
}