SVTRv2X: Enhanced scene text recognition via self-distilled mixture-of-experts
Jian Guo, Hanxin Cui, Wengang Tang, Xuehai Zhou, Xing Xu, Qianqian Cheng · 2026
Read SVTRv2X: Enhanced scene text recognition via self-distilled mixture-of-experts by Jian Guo, Hanxin Cui, Wengang Tang, Xuehai Zhou, Xing Xu, Qianqian Cheng (2026) 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/plos-pone-0349085.txt.
Description
Jian Guo and co-authors upgraded a scene text recognition system in 2026, the technology that reads signs and labels in photos for cars and apps. Light models falter on rotated, warped, or oddly styled text, a common real world case.
Three additions drive the upgrade. A jumbling step shuffles image patches to break the model's reliance on fixed layout. Self teaching passes knowledge from deep layers to shallow ones. A mixture of expert paths handles varied styles. Together they aim at robustness without a heavy model. The design accepts that street text is unruly and trains for that directly.
Catalog record
| CREATOR | Jian Guo, Hanxin Cui, Wengang Tang, Xuehai Zhou, Xing Xu, Qianqian Cheng |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46667 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |