README.md

metadata

language:
  - en
task_categories:
  - document-question-answering
  - optical-character-recognition
  - information-extraction
license: other

Receipt Dataset — Standardized & AI-Assisted Annotations

This dataset contains 2800 receipt images with paired JSON annotations, curated and standardized from multiple publicly available receipt datasets.

Structure

• receipts/ — receipt images
• annotations/ — JSON annotations (paired by filename)
• MANIFEST.csv — sample index
• NOTICE.md — attribution and methodology
• licenses/ — upstream license texts

Annotation methodology

Annotations were created using:

• original dataset annotations (where available)
• AI-assisted OCR and information extraction using commercial AI models
• manual review and normalization

Annotations are model-assisted labels intended for training and evaluation.

Source breakdown

sroie: 971 zenodo: 813 cord: 798 express: 198

Full dataset download

The full dataset is provided as a single archive:

• receipt_dataset_2780.tgz (contains receipts/ and annotations/)
• receipt_dataset_2780.tgz.sha256 (checksum)

Extract:

tar -xzf receipt_dataset_2780.tgz

## Licensing

This dataset redistributes upstream datasets under their original licenses.
See `NOTICE.md` and `licenses/` for full details.