DocLayNet
DocLayNet: A Large Human-Annotated Dataset for Document-Layout Analysis
Install / Use
/learn @DS4SD/DocLayNetREADME
DocLayNet
DocLayNet is a human-annotated document layout segmentation dataset containing 80863 pages from a broad variety of document sources.
News
| | | | :---------- | :----------------------------------- | | Jan. 26th, 2023 | The DocLayNet dataset is available on Hugging Face ds4sd/DocLayNet. | | Jan. 13th, 2023 | We are hosting a competition on layout segmentation in corporate documents in ICDAR 2023. Find the details in the competition website. |
Overview
DocLayNet provides page-by-page layout segmentation ground-truth using bounding-boxes for 11 distinct class labels on 80863 unique pages from 6 document categories. It provides several unique features compared to related work such as PubLayNet or DocBank:
- Human Annotation: DocLayNet is hand-annotated by well-trained experts, providing a gold-standard in layout segmentation through human recognition and interpretation of each page layout
- Large layout variability: DocLayNet includes diverse and complex layouts from a large variety of public sources in Finance, Science, Patents, Tenders, Law texts and Manuals
- Detailed label set: DocLayNet defines 11 class labels to distinguish layout features in high detail.
- Redundant annotations: A fraction of the pages in DocLayNet are double- or triple-annotated, allowing to estimate annotation uncertainty and an upper-bound of achievable prediction accuracy with ML models
- Pre-defined train- test- and validation-sets: DocLayNet provides fixed sets for each to ensure proportional representation of the class-labels and avoid leakage of unique layout styles across the sets.
Using the dataset with :hugs: Hugging Face
The DocLayNet dataset is available on Hugging Face at ds4sd/DocLayNet.
>>> from datasets import load_dataset
>>> dataset = load_dataset("ds4sd/DocLayNet")
>>> dataset
DatasetDict({
train: Dataset({
features: ['image_id', 'image', 'width', 'height', 'doc_category', 'collection', 'doc_name', 'page_no', 'objects'],
num_rows: 69375
})
validation: Dataset({
features: ['image_id', 'image', 'width', 'height', 'doc_category', 'collection', 'doc_name', 'page_no', 'objects'],
num_rows: 6489
})
test: Dataset({
features: ['image_id', 'image', 'width', 'height', 'doc_category', 'collection', 'doc_name', 'page_no', 'objects'],
num_rows: 4999
})
})
Download
| Dataset | Records | Size(GB) | URL | |------------------|---------|----------|-------------| | DocLayNet core dataset| 80,863 | 28 GiB | Download| | DocLayNet extra files| 80,863 | 7.5 GiB | Download|
Additionally, we provide the labeling guideline used for training of the annotation experts here.
Dataset structure
DocLayNet provides four types of data assets:
- PNG images of all pages, resized to square
1025 x 1025px - Bounding-box annotations in COCO format for each PNG image
- Extra: Single-page PDF files matching each PNG image
- Extra: JSON file matching each PDF page, which provides the digital text cells with coordinates and content
The dataset is organized in the following directory structure:
Doclaynet core dataset
├── COCO
│ ├── test.json
│ ├── train.json
│ └── val.json
├── PNG
│ ├── <hash>.png
│ ├── ...
Doclaynet extra files
├── PDF
│ ├── <hash>.pdf
│ ├── ...
├── JSON
│ ├── <hash>.json
│ ├── ...
Data format details
Pages
Document pages are provided as bitmap images (PNG) and in original PDF format. The layout annotations in COCO format are referring to the PNG images only.
COCO annotations
For bounding-box annotations, DocLayNet provides standard COCO format annotations as defined here. Each COCO image record contains additional custom fields to allow data sub-selection and provide provenance.
Example page with bounding-box annotations
Example COCO image record
...
{
"id": 1,
"width": 1025,
"height": 1025,
"file_name": "132a855ee8b23533d8ae69af0049c038171a06ddfcac892c3c6d7e6b4091c642.png",
// Custom fields:
"doc_category": "financial_reports" // high-level document category
"collection": "ann_reports_00_04_fancy", // sub-collection name
"doc_name": "NASDAQ_FFIN_2002.pdf", // original document filename
"page_no": 9, // page number in original document
"precedence": 0, // Annotation order, non-zero in case of redundant double- or triple-annotation
},
...
The doc_category field uses one of the following constants:
financial_reports,
scientific_articles,
laws_and_regulations,
government_tenders,
manuals,
patents
Example COCO annotation records
The annotation records shown below all belong to the page image shown above. Every bounding-box is a separate record, matched to the common image_id.
"annotations": [
{
"id": 8,
"image_id": 1,
"category_id": 1,
"bbox": [
210.06018382352943,
31.14536268939389,
173.9850743464052,
39.270946654040586
],
"segmentation": [
[
210.06018382352943,
31.14536268939389,
210.06018382352943,
70.41630934343448,
384.04525816993464,
70.41630934343448,
384.04525816993464,
31.14536268939389
]
],
"area": 6832.558573256964,
"iscrowd": 0,
"precedence": 0
},
{
"id": 9,
"image_id": 1,
"category_id": 7,
"bbox": [
434.9334063800317,
-0.4906348977078778,
589.8585905504372,
590.2337819428021
],
"segmentation": [
[
434.9334063800317,
-0.4906348977078778,
434.9334063800317,
589.7431470450942,
1024.791996930469,
589.7431470450942,
1024.791996930469,
-0.4906348977078778
]
],
"area": 348154.46671203536,
"iscrowd": 0,
"precedence": 0
},
{
"id": 10,
"image_id": 1,
"category_id": 8,
"bbox": [
66.99346405228758,
112.10344760101009,
290.869358251634,
13.66279703282828
],
"segmentation": [
[
66.99346405228758,
112.10344760101009,
66.99346405228758,
125.76624463383837,
357.8628223039216,
125.76624463383837,
357.8628223039216,
112.10344760101009
]
],
"area": 3974.0890048610913,
"iscrowd": 0,
"precedence": 0
},
{
"id": 11,
"image_id": 1,
"category_id": 10,
"bbox": [
66.99346405228758,
133.5865287247475,
325.3560694444444,
131.31064046717177
],
"segmentation": [
[
66.99346405228758,
133.5865287247475,
66.99346405228758,
264.89716919191926,
392.34953349673196,
264.89716919191926,
392.34953349673196,
133.5865287247475
]
],
"area": 42722.71385863161,
"iscrowd": 0,
"precedence": 0
},
{
"id": 12,
"image_id": 1,
"category_id": 10,
"bbox": [
66.99346405228758,
272.84557828282834,
325.4857017973857,
131.3025
],
"segmentation": [
[
66.99346405228758,
272.84557828282834,
66.99346405228758,
404.14807828282835,
392.4791658496732,
404.14807828282835,
392.4791658496732,
272.84557828282834
]
],
"area": 42737.08636025123,
"iscrowd": 0,
"precedence": 0
},
{
"id": 13,
"image_id": 1,
"category_id": 10,
"bbox": [
66.99346405228758,
414.8919160353536,
325.69678145424837,
80.83059406565656
],
"segmentation": [
[
66.99346405228758,
414.8919160353536,
66.99346405228758,
495.72251010101013,
392.6902455065359,
495.72251010101013,
392.6902455065359,
414.8919160353536
]
],
"area": 26326.26433021921,
"iscrowd": 0,
"precedence": 0
},
{
"id": 14,
"image_id": 1,
"category_id": 10,
"bbox": [
112.37566899509804,
626.7556887626263,
863.111772998366,
310.9754294507576
],
"segmentation": [
[
112.37566899509804,
626.7556887626263,
112.37566899509804,
937.7311182133839,
975.487441993464,
937.7311182133839,
975.487441993464,
626.7556887626263
]
],
"area": 268406.55427217163,
"iscrowd": 0,
"precedence": 0
},
{
"id": 15,
"image_id": 1,
"category_id": 5,
"bbox": [
18.32874183006536,
1005.2406660353536,
7.4496732026143775,
10.353535094696968
],
"segmentation": [
[
18.32874183006536,
1005.2406660353536,
18.32874183006536,
1015.5942011300506,
25.77841503267974,
1015.5942011300506,
25.77841503267974,
1005.
Security Score
Audited on Mar 24, 2026