[Synthetic] Medical Documents Collection On HF (for Testing Raster Processing Pipelines)

I’ve created a dataset (over 7,700 files/17GB) on Hugging Face that can help with testing and tuning scanned document processing:

https://huggingface.co/datasets/morzel85/synthetic-medical-document-recognition-benchmark

It’s meant to be realistic but fully synthetic (based on Synthea/FHIR).

I used it to assess OCR + NER for de-identification, and because documents are clearly marked as synthetic these are safe for demos. The README file contains all details.

Maybe it will be useful for someone else. Released under cc-by-4.0 (free for any use, watch out: some datasets are produced with components that not allow commercial use).