ADOPD is a large-scale dataset for document page decomposition. It introduces a novel data-driven taxonomy discovery approach, combining large-scale pretrained models with human-in-the-loop validation to ensure diversity and balance. ADOPD provides densely annotated document images for four key tasks: Doc2Mask, Doc2Box, Doc2Tag, and Doc2Seq. Each image includes entity masks, text bounding boxes, and descriptive tags and captions. Among these tasks, Doc2Mask and Doc2Box are annotated by humans with model assistance, while Doc2Tag and Doc2Seq are initially generated by large language models (LLMs) and subsequently refined by human annotators.
Jiuxiang Gu, Xiangxi Shi, Jason Kuen, Lu Qi, Ruiyi Zhang, Anqi Liu, Ani Nenkova, Tong Sun
The dataset annotations can be downloaded here.
Images can be downloaded using the image source entries and/or URLs provided in the json file or urls.csv. Please note that we do not own the copyright of the images. It is solely your responsibility to check the original licenses of the images before using them. Any use of the images is at your own discretion and risk.
Please cite our ICLR 2024 paper if you use the ADOPD dataset in your work:
@inproceedings{
gu2024adopd,
title={{ADOPD}: 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}
}