1,000 scanned receipt images across text localization and information extraction categories. The dataset provides ground truth labels for company, address, date, and total amount alongside raw OCR transcriptions.
Use Cases
- Train a key information extraction (KIE) model using the 'company' and 'total' labels
- Develop optical character recognition (OCR) systems using the provided bounding box coordinates and text transcriptions
- Fine-tune layout-aware machine learning models using the spatial relationship between text blocks
Strengths
- 1,000 scanned receipt images in JPG format
- Ground truth labels for 'company', 'date', 'address', and 'total' fields
- Bounding box coordinates for text localization provided in text files