receipts1

Receipts1 is a dataset and accompanying software framework designed for research in receipt image understanding and information extraction. It provides a large collection of scanned receipts together with structured annotations and tools to support optical character recognition, layout analysis, and data extraction tasks in a reproducible way.

The dataset comprises thousands of receipt images drawn from diverse retailers and regions, including both real and synthetic content. Annotations include the transcription of text, bounding boxes for key regions, and structured fields such as merchant name, date, total amount, tax, and currency. Line-item details identify individual products, quantities, prices, and subtotals.

Annotation is produced via a combination of automated labeling and human verification, with quality assurance metrics reported for OCR accuracy and field extraction F1 scores. The dataset also provides baseline models and evaluation scripts to facilitate comparison across research efforts.

Typical applications include training and benchmarking OCR systems, information extraction pipelines, receipt parsing, and cross-domain transfer learning for related document types such as invoices and bills.

Access is provided through an open data repository and accompanying software library. Documentation covers data formats, licensing terms intended to enable research and education, and guidance on citation. The project welcomes community contributions and periodic updates, and it is designed to integrate with common machine learning workflows and evaluation benchmarks.