A live reading and discussion of the RT-DocLayout paper with Isaac Flath and Joe Barrow, exploring PP-DocLayoutV3’s boxes, reading order, and segmentation.

Document Layout with PP-DocLayoutV3

Why a rectangle can include neighboring text, and how PP-DocLayoutV3 follows the shape of a region.

PP-DocLayoutV3 is the layout stage of an OCR pipeline and the curved outlines are segmentation boundaries. The paper tests document images, but there are many other use cases for this feature

Curved text and segmentation

How the model compares pairs of blocks to order a page.

PP-DocLayoutV3 finds document regions and predicts reading order together. It compares pairs of regions in parallel, estimates how many regions come before each one, and sorts them into a sequence. The paper’s experiment found learning these tasks together reduces reading-order errors compared with training the ordering model separately.

Putting blocks in reading order

Joining tables and paragraphs across pages, organizing headings, and deciding when the extra processing helps.

Research

Upcoming

Document Layout with PP-DocLayoutV3

Past talks

OCR Post-Processing and MinerU-Popo