Hawkeye: Reading Illegible Text with Vision Language Models

Brian Tang, Kang G. Shin

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 tha…