Metraipolis › archive › dc_b2fce30fdcc7

Hawkeye: Reading Illegible Text with Vision Language Models — arXiv:2610.41254

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.41254
License (as archived)
CC0-1.0
Archived
2026-11-03T05:46:19.390Z
SHA-256 of the archived text
8ac6bcb06596b8bbdbe64d139d69044484c11e8d1e4cbfe4935161ebc7943c58
Stable URL of this page
https://metraipolis.com/documento/dc_b2fce30fdcc7

How to cite

Hawkeye: Reading Illegible Text with Vision Language Models — arXiv:2610.41254 — arXiv (en), https://arxiv.org/abs/2610.41254 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_b2fce30fdcc7 (Metraipolis archive dc_b2fce30fdcc7, sha256 8ac6bcb06596b8bbdbe64d139d69044484c11e8d1e4cbfe4935161ebc7943c58)
BibTeX
@misc{dcb2fce30fdcc7,
  title = {Hawkeye: Reading Illegible Text with Vision Language Models — arXiv:2610.41254},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.41254},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_b2fce30fdcc7; sha256 8ac6bcb06596b8bbdbe64d139d69044484c11e8d1e4cbfe4935161ebc7943c58},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.41254. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

While modern Vision-Language Models (VLMs) have been remarkably successful in general scene text recognition, they consistently struggle with the long-tail distribution of text that is degraded by motion blur or low resolution. These limitations are particularly apparent in egocentric applications like smart glasses, where text is often captured in noisy environments. In this paper, we introduce the \textbf{BSTRB} (Blurry Scene Text Recognition Benchmark), a novel benchmark of 8,398 noisy image samples with high-fidelity annotations derived from 1,699 real-world video sequences. To support this, we present \textbf{Hawkeye}, a VLM fine-tuned on BSTRB to read and reason about illegible text by leveraging contextual priors and pixel-level alignment. Our results show that \name matches much larger VLMs on BSTRB overall and exceeds them on text at or below 20 pixels, in a single pass suited to real-time deployment. We characterize the resolution at which current VLMs begin to fail.

Tema degli esploratori: statistica e scienza dei dati (elenco arXiv cs.CV, nuovi lavori del giorno).
Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2610.41254 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-03T05:33:32.283Z dalla pagina https://arxiv.org/list/cs.CV/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).