Beyond Pixels: Text Enhances Generalization in Real-World Image Restoration

TL;DR

This paper introduces a novel approach that uses text as an auxiliary invariant to improve the generalization of diffusion-based image restoration models on real-world data, addressing their limitations with out-of-distribution images.

Contribution

The paper proposes Res-Captioner, a new module that generates tailored textual descriptions to enhance model robustness, and introduces RealIR, a benchmark for diverse real-world scenarios.

Findings

Res-Captioner improves generalization of diffusion models.

Text as an auxiliary invariant enhances restoration performance.

Res-Captioner is fully plug-and-play.

Abstract

Generalization has long been a central challenge in real-world image restoration. While recent diffusion-based restoration methods, which leverage generative priors from text-to-image models, have made progress in recovering more realistic details, they still encounter "generative capability deactivation" when applied to out-of-distribution real-world data. To address this, we propose using text as an auxiliary invariant representation to reactivate the generative capabilities of these models. We begin by identifying two key properties of text input: richness and relevance, and examine their respective influence on model performance. Building on these insights, we introduce Res-Captioner, a module that generates enhanced textual descriptions tailored to image content and degradation levels, effectively mitigating response failures. Additionally, we present RealIR, a new benchmark designed to capture diverse real-world scenarios. Extensive experiments demonstrate that Res-Captioner significantly enhances the generalization abilities of diffusion-based restoration models, while remaining fully plug-and-play.