OCR Post-Processing and MinerU-Popo
OCR often makes errors because of a lack of cross page context. Many models and pipelines treat each page independently, but that's not how actual PDFs work. For examples tables split across pages and headings and sections span several pages. If you don't have the prior page context then there are some things you cannot OCR correctly.
When you want to search, extract, and use PDFs in a product, workflow, or RAG system this context is important.
That's why Joe Barrow and I are discussing MinerU-Popo: Universal Post-Processing Model for Structured Document Parsing live this Wednesday.
We’ll use the paper to talk through:
What needs to be fixed after OCR has extracted text
How MinerU-Popo handles paragraphs and tables across pages, organizes headings, and images with text.
What situations this extra processing helps with and what types of workflows this applies (and doesn't) apply to
Wednesday, October 14 · 7:30–8:30 p.m. Eastern
It’s free. Sign up here to get the calendar invitation + recording & notes after emailed to you.