← Podcast

Podcast · Past conversation

OCR Post-Processing and MinerU-Popo

With Isaac Flath and Joe Barrow

OCR Post-Processing and MinerU-Popo, with Isaac Flath and Joe Barrow

This talk has ended. The recording hasn’t been added here yet.

A table can continue onto the next page, and a section can span several pages. When an OCR pipeline processes each page independently, it can miss those connections.

Joe Barrow and I are reading MinerU-Popo and discussing what happens after individual pages have been parsed. We’ll look at how it uses context across pages, and where we would use it for document search and extraction.

What we discussed

  • Joining tables and paragraphs that continue onto the next page
  • Organizing headings into a document structure
  • When the extra processing helps, and when it isn’t needed

Read the MinerU-Popo paper ↗