1,000 scanned receipt images paired with OCR transcriptions and structured entity labels. The data covers four key information extraction categories: company, date, address, and total amount.
Use Cases
- Train document AI models to perform key-value pair extraction using the 'company' and 'total' labels
- Benchmark OCR engines on text detection using the 8-point coordinate bounding boxes
- Fine-tune multimodal models that integrate visual features with text tokens for layout analysis
Strengths
- 1,000 scanned receipt images provided in JPG format
- OCR ground truth includes 8-point polygon coordinates (x1, y1 to x4, y4) for precise text localization
- Information extraction labels are provided in JSON format for 'company', 'date', 'address', and 'total' keys