Dataset Viewer

The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

Khmer Document Corpus

Overview

Khmer Document Corpus is an open-source dataset designed for Document AI research, focusing on Khmer, English, and multilingual Khmer-English documents.

The long-term goal of this project is to build one of the largest open document corpora for Khmer, supporting research in document understanding, OCR, document layout analysis, table extraction, font recognition, and PDF-to-Word reconstruction.

Version v0.1 is the initial release containing raw PDF documents and metadata.


Objectives

This dataset is being developed to support future research in:

  • Document AI
  • Optical Character Recognition (OCR)
  • Khmer OCR
  • English OCR
  • Multilingual OCR
  • Document Layout Analysis
  • Table Recognition
  • Reading Order Detection
  • Font Recognition
  • PDF Parsing
  • PDF-to-Word Conversion
  • Vision-Language Models (VLM)
  • Information Extraction
  • Large Multimodal Models (LMM)

Languages

The dataset currently focuses on:

  • Khmer (km)
  • English (en)
  • Mixed Khmer-English

Future versions may include additional Southeast Asian languages.


Document Types

The corpus will gradually include publicly available documents such as:

  • Government Reports
  • Government Forms
  • Laws
  • Gazettes
  • Books
  • Research Papers
  • Manuals
  • Annual Reports
  • Financial Reports
  • Certificates
  • Contracts
  • Invoices
  • Receipts
  • Newspapers
  • Magazines
  • Presentations

Current Release

Version

v0.1

Current contents:

  • Raw PDF documents
  • Basic document metadata

Future releases will include:

  • Page preview images
  • OCR annotations
  • Layout annotations
  • Table annotations
  • Font annotations
  • Reading order annotations
  • Ground truth DOCX files
  • Structured JSON annotations

Dataset Structure

khmer-document-corpus/

├── metadata/
├── pdf/
├── preview/
├── scripts/
└── README.md

Metadata

Each document may contain the following metadata.

Field Description
id Unique document ID
language Document language
category Document category
pages Number of pages
native_pdf Whether text is embedded
scanned Whether OCR is required
has_tables Contains tables
has_images Contains images
has_header Header detected
has_footer Footer detected
source Collection source
license Original document license

Intended Use

This dataset is intended for:

  • Academic research
  • OCR benchmarking
  • Document AI
  • Computer Vision
  • Natural Language Processing
  • Vision-Language Models
  • PDF processing
  • Information Extraction
  • Machine Learning

Data Collection

Documents are collected from publicly available sources or documents that can be legally redistributed.

Potential sources include:

  • Government publications
  • Universities
  • Public reports
  • Research publications
  • Open-access books
  • Public forms
  • Public documentation

Documents with unclear redistribution rights should not be included.


Licensing

This repository is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).

The repository license applies to the dataset organization and metadata.

Individual documents may have their own licenses or terms of use. Users are responsible for complying with the license associated with each original document.


Limitations

Current limitations include:

  • Small initial dataset
  • No OCR annotations
  • No layout annotations
  • No font annotations
  • No table annotations
  • No reading-order annotations

These will be added in future releases.


Roadmap

Version Status Description
v0.1 Raw PDF Corpus
v0.2 Metadata Expansion
v0.3 Page Preview Images
v0.4 OCR Ground Truth
v0.5 Layout Annotations
v0.6 Table Annotations
v0.7 Font Annotations
v0.8 Reading Order
v1.0 PDF-to-Word Benchmark

Citation

If you use this dataset in your research, please cite:

@dataset{khmer_document_corpus_2026,
  title={Khmer Document Corpus},
  author={Darachhat},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/Darachhat/khmer-document-corpus}
}

Contributing

Contributions are welcome.

Future contributions may include:

  • Public PDF collections
  • Metadata improvements
  • OCR annotations
  • Layout annotations
  • Table annotations
  • Font annotations
  • Documentation improvements

Please ensure that all contributed documents can be legally redistributed before submitting a pull request.


Contact

Maintainer: Darachhat

For questions, suggestions, or collaboration, please open an issue in the repository.

Downloads last month
-