v1.7.5
datalab-to/markerv1.7.5Jun 11, 2025by VikParuchuri
AI Summary
Introduces a block type relabeling processor and fixes bugs related to rotated text and equation rendering in Surya.
Key Highlights
- Added `--block_relabel_str` processor to force relabeling based on rules and confidence thresholds
- Fixed rotated text causing words to be combined without spaces
- Fixed Surya inserting `<br>` tags into equations
New Features
- Block relabel processor
- Bugfixes for rotated text and equation rendering
Full Release Notes
## Bugfixes - In certain PDFs (<1%), rotated text caused words to be combined without spaces (like `Helloworld`). This was patched in pdftext, with the fix pulled in. - Surya sometimes inserted `<br>` tags into equations - this was fixed in surya ## Block type rules Add a processor to force relabel blocks based on rules. An example is `--block_relabel_str "Table:Picture:0.97,Form:Picture:1.0,TableOfContents:Picture:0.97"`. Each comma separated block is a rule, with the original block type first, the new block type second, and the confidence threshold last. The original block will be relabeled to the new block type if the layout model confidence is less than the threshold. ## What's Changed * Add block relabel processor by @tarun-menta in https://github.com/datalab-to/marker/pull/747 * Dev by @VikParuchuri in https://github.com/datalab-to/marker/pull/733 * Version bump by @VikParuchuri in https://github.com/datalab-to/marker/pull/748 **Full Changelog**: https://github.com/datalab-to/marker/compare/v1.7.4...v1.7.5