Dataset · Voxel51
Synthetic invoice images for OCR
Voxel51/high-quality-invoice-images-for-ocr
Over 8,000 synthetic English invoice scans partially annotated for OCR and document parsing.
What's inside
The dataset from Voxel51 contains 8,181 synthetic English invoice images prepared for use in the FiftyOne library. 1,489 images have full annotations with structured JSON and raw OCR text, while the remaining 6,692 files are unannotated. It is suitable for optical character recognition and document structure understanding tasks.
What you can do with it
- Training and fine-tuning OCR models on financial documents
- Training the extraction of structured JSON fields from invoices
- Experiments with semi-supervised learning on unannotated data
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 27.9K
- Likes
- 9
How to load it
from datasets import load_dataset
ds = load_dataset("Voxel51/high-quality-invoice-images-for-ocr")