TiCLS : Tightly Coupled Language Text Spotter
Leeje Jang, Yijun Lin, Yao-Yi Chiang, Jerod Weinman · 2026
Read TiCLS : Tightly Coupled Language Text Spotter by Leeje Jang, Yijun Lin, Yao-Yi Chiang, Jerod Weinman (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/arxiv-2602-04030.txt.
Description
Text in the wild is often broken, tilted, or half-hidden. Leeje Jang and 3 co-authors built TiCLS, a scene text spotter that reads with linguistic help. Their 2026 system tightly couples detection and recognition with a character-level pretrained language model.
A linguistic decoder fuses visual features with language knowledge and can start from the pretrained model's weights. Ambiguous fragments get resolved by what characters plausibly follow. On ICDAR 2015 and Total-Text benchmarks, TiCLS reached state-of-the-art results. The lesson generalizes: perception systems read better when they know the language they are looking at.
Catalog record
| CREATOR | Leeje Jang, Yijun Lin, Yao-Yi Chiang, Jerod Weinman |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 37 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-04030