<ul data-eligibleForWebStory="true"><li>Image restoration methods often struggle to reconstruct textual regions accurately, resulting in text-image hallucination.</li><li>Text-Aware Image Restoration (TAIR) is introduced to recover visual contents and textual fidelity simultaneously.</li><li>SA-Text, a large-scale benchmark of scene images annotated with text instances, is presented.</li><li>A multi-task diffusion framework called TeReDiff integrates features from diffusion models into a text-spotting module.</li><li>The joint training of components allows for rich text representations used in denoising.</li><li>Experiments show that the approach outperforms existing methods, improving text recognition accuracy.</li><li>Project page: https://cvlab-kaist.github.io/TAIR/</li></ul>

Text-Aware Image Restoration with Diffusion Models

Discover more