metadata
license: cc-by-4.0
Description
The CORD (Consolidated Receipt Dataset) dataset contains receipts annotated for key information extraction. It was released for the 2019 ICDAR competition on scanned receipts.
Content
- 1,000 receipts (800 train/ 100 val/ 100 test)
- Entities include menu items, totals, store information, and dates
- OCR text + layout information available
- More fine-grained annotations than in SROIE (e.g. line items in receipts)
- Useful for benchmarking models on dense receipt parsing
Intended Use
- NER/KIE on receipts
License
Provided by clovaai.