Extend-AI/RealDoc-Bench-Layout

1---2license: cc-by-4.03task_categories:4  - object-detection5tags:6  - document-layout7  - layout-analysis8  - coco9size_categories:10  - 1K<n<10K11configs:12  - config_name: default13    data_files:14      - split: test15        path: manifest.csv16---17 18# RealDocBench-Layout19 20A 1,500-page document-layout benchmark for evaluating layout-detection21models on real-world documents. COCO-style annotations across **9 block22classes**.23 24## Contents25 26- `images/` — 1,500 page images (PNG / JPG / occasional WebP-as-PNG; see Caveats).27- `annotations/<pageId>.json` — per-page COCO files, each with a single28  `image` record, an `annotations` list, a `categories` list, and a29  `page_info` block.30- `manifest.csv` — pageId → domain + source URLs. The canonical row list.31 32## Block classes (9)33 34```35text, heading, section_heading, header, footer, page_number,36figure, table, key_value37```38 39| class | what it covers |40|---|---|41| `text` | body text and other inline content |42| `heading` | document title |43| `section_heading` | section / subsection titles |44| `header` | running page header |45| `footer` | running page footer |46| `page_number` | page-number indicators |47| `figure` | images, diagrams, charts, barcodes, QR codes |48| `table` | tabular data |49| `key_value` | form fields / key-value pairs |50 51The companion benchmark loader at52[`extend-hq/realdoc-bench`](https://github.com/extend-hq/realdoc-bench)53walks each annotation file and produces 9-class block predictions; see54`realdoc_bench/layout/normalizers/coco.py`.55 56## Loading57 58```python59from huggingface_hub import snapshot_download60path = snapshot_download(repo_id="Extend-AI/RealDocBench-Layout", repo_type="dataset")61```62 63## License64 65Annotations: CC-BY-4.0. Page images carry their original per-source66licenses; the manifest's `sourceUrl` column points back to each one.67 68### Takedown and Opt-out requests69If you are the owner of a document included in this dataset and would like it removed, you can submit a takedown request. We process requests on a monthly cadence and maintain a public manifest of removed records at takedowns/removed_ids.jsonl.70 71To request removal, open an issue in the Community tab or email [support@extend.ai] with:72 73A document identifier (filename, URL, or row ID if known), and the scope of removal (specific rows, an entire source document, or all documents from a given source)74