ViTSTR-Transducer: Cross-Attention-Free Vision Transformer Transducer for Scene Text Recognition
Rina Buoy, Masakazu Iwamura, Sovila Srun, Koichi Kise, Georgina Cosma · 2023
Read ViTSTR-Transducer: Cross-Attention-Free Vision Transformer Transducer for Scene Text Recognition by Rina Buoy, Masakazu Iwamura, Sovila Srun, Koichi Kise, Georgina Cosma (2023) 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/pmc-10744502.txt.
Description
Rina Buoy and co-authors present a leaner system for reading text in photos, in a 2023 paper. Systems that recognize scene text usually align image and language features with cross-attention, a step that costs heavy computation and slows things down on modest hardware. Their design drops cross-attention entirely while still learning an internal sense of language.
The system pairs a vision transformer front end with a transducer borrowed from speech recognition. In tests, it needed far fewer decoding operations and ran faster than attention-based rivals, while matching their accuracy. It also beat a simpler context-free baseline and held its own against recent leading methods. The authors target use in low-resource settings.
Catalog record
| CREATOR | Rina Buoy, Masakazu Iwamura, Sovila Srun, Koichi Kise, Georgina Cosma |
|---|---|
| YEAR | 2023 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |