AI Briefing5,000 Australian Medical PDFs Released
Key point
5,000 synthetic document-style medical PDFs from Australia have been released, along with a free 50-document sample.
Details
5,000 synthetic PDFs have been released to target the gap in Australian medical document training data.
They were generated using a reproducible Python pipeline that follows NSW Health hospital form and GP clinic document styles, with all entities being fake. Each document includes the text "SYNTHETIC TRAINING DOCUMENT - NOT FOR CLINICAL USE" at the bottom.
- 45 document types: discharge summaries, emergency department assessments, referral letters, pathology reports, prescriptions, mental health assessments, anesthesia records, etc.
- 4 quality tiers: clean, scanned, poor, fax
- Ground truth data: field-level
(x, y, w, h, page)coordinates provided viabboxes_jsoninground_truth.csvand per-documentbboxes.jsonl - Free sample: 50 documents, 29 types, 682 bbox annotations, with 27 scanned, 16 clean, 6 poor, and 1 fax
It's designed to be used right away for training document VLMs and OCR models like LayoutLMv3, Donut, and DocFormer, and is licensed under CC-BY-NC 4.0.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Class Action Suit Alleges OpenAI Routed ChatGPT Conversations to Human Reviewers Without Clear DisclosureOpenAI Tracks User Behavior on External Websites via Cookies Linked to ChatGPT Accounts