11,000 receipt images annotated with OCR text and hierarchical semantic labels across 30 distinct classes. The dataset provides quadrilateral bounding boxes and text strings for every identified element, organized into a structured format for document intelligence tasks.
Use Cases
- Train a named-entity recognition (NER) model to classify text segments into 'label' categories such as 'total.cashprice'
- Develop document layout analysis tools using the 'quad' coordinates to understand spatial relationships between receipt items
- Fine-tune multimodal models to extract structured data from 'text' and image features for automated expense reporting
Strengths
- 11,000 high-resolution receipt images with corresponding JSON annotation files
- Hierarchical labels including specific sub-fields like 'menu.nm', 'menu.cnt', and 'total.total_price'
- Quadrilateral coordinates ('quad') for precise text localization on warped or tilted physical receipts
- 30 semantic categories covering store information, line items, and payment details