|
|
--- |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
**Description** |
|
|
The **SROIE** (Scanned Receipts OCR and Information Extraction) dataset consists of scanned receipts collected for the ICDAR 2019 SROIE challenge. It is designed for NER/key information extraction from receipts. |
|
|
|
|
|
**Content** |
|
|
|
|
|
- 973 annotated receipts (626 train/347 test) |
|
|
- 4 entity types: _company, date, address, total_ |
|
|
- OCR text + layout information available |
|
|
- Short documents (receipts) with sparse (majority of tokens 'O' outside a mention) but critical fields |
|
|
- Widely used as a benchmark for document information extraction |
|
|
|
|
|
|
|
|
**Intended Use** |
|
|
|
|
|
- NER/KIE on receipts |
|
|
|
|
|
|
|
|
**License** |
|
|
[Original dataset](https://rrc.cvc.uab.es/?ch=13) provided by ICDAR 2019 SROIE organizers. |
