adopd2024

# ADOPD: A Large-Scale Document Page Decomposition Dataset [](.top_ADOPD_Tasks.png) **ADOPD** is a large-scale dataset designed for document image understanding. It introduces a novel data-driven document taxonomy discovery framework that combines large-scale pretrained models with a human-in-the-loop refinement process. The dataset supports four core tasks and includes rich annotations to foster progress in document analysis. --- ## 📊 Dataset Summary - **Total Images**: 120,000 - **Languages**: - English: 60,000 - Chinese: 20,000 - Japanese: 20,000 - Others: 20,000 ADOPD ensures **diversity and balance** across document types and languages, validated through extensive experiments. --- ## 🧩 Supported Tasks | Task | Description | Field Name | |------------|--------------------------------------------|----------------------| | **Doc2Mask** | Segment entity regions as pixel-level masks | `Entity_Masks` | | **Doc2Box** | Detect and group OCR text blocks | `Grouped_OCR_Blocks` | | **Doc2Tag** | Predict high-level semantic tags | `Tags` | | **Doc2Seq** | Generate abstracted captions | `Caption` | --- ## 📝 Annotation Fields Each data sample includes the following fields: - `ID` / `URL_sha256`: Unique identifier for each document image - `URL`: Direct link to download the image - `Height`, `Width`: Image resolution in pixels - `Entity_Masks`: Human-annotated segmentation masks for document entities - `Grouped_OCR_Blocks`: Grouped OCR text blocks with bounding boxes - `Caption`: Human-written descriptive caption summarizing the content - `Tags`: Predicted document-level semantic tags - `OCR_Text`: Raw plain-text extracted from the image - `Language`: Language of the document content - `NSFW`: Indicator flag for not-safe-for-work (NSFW) content --- ## 📄 Citation If you use ADOPD in your research, please cite: ```bibtex @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} } ``` --- ## 📜 License - **License**: [CC BY-NC-ND 4.0](https://creativecommons.org/licenses/by-nc-nd/4.0/) - For **non-commercial use only**. Redistribution and derivative works are **not permitted**.