Spatially-Grounded Document Retrieval via Patch-to-Region Relevance Propagation
Athos Georgiou · 2025
Read Spatially-Grounded Document Retrieval via Patch-to-Region Relevance Propagation by Athos Georgiou (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-02660.txt.
Description
Athos Georgiou improves document search so answers point to the right region, not just the right page. Late interaction models like ColPali embed each page as an image and match query words to visual patches closely. Page level results, though, give generation systems more text than they need.
Text extraction tools give exact boxes but judge relevance weakly. Georgiou propagates patch relevance into regions, joining fine matching with precise location. Region level retrieval could make question answering over manuals, forms, and reports sharper, cheaper, and easier for users to verify by eye. A hit at the region scale gives the writer the lines it needs, not the whole file to wade through.
Catalog record
| CREATOR | Athos Georgiou |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-02660