UniCoRN: Latent Diffusion-based Unified Controllable Image Restoration Network across Multiple Degradations

TL;DR

UniCoRN tackles the challenge of restoring images corrupted by multiple, unknown degradations by embedding low-level restoration cues into a latent diffusion framework via a multi-head control network and mixture-of-experts adaptation. It introduces a Multi-Level Condition Network to fuse primary and secondary cues, a task-aware MOE adapter, and a Task Stabilizer Unit to enable stable multi-task learning under a curriculum that progresses from single to mixed degradations. The approach achieves state-of-the-art performance on the MetaRestore metalens benchmark and shows strong generalization to unseen degradations, while offering competitive inference speed. These contributions demonstrate the potential for robust, universal image restoration in real-world, multi-degradation scenarios without requiring degradation-specific priors, with implications for diverse vision applications and imaging pipelines.

Abstract

Image restoration is essential for enhancing degraded images across computer vision tasks. However, most existing methods address only a single type of degradation (e.g., blur, noise, or haze) at a time, limiting their real-world applicability where multiple degradations often occur simultaneously. In this paper, we propose UniCoRN, a unified image restoration approach capable of handling multiple degradation types simultaneously using a multi-head diffusion model. Specifically, we uncover the potential of low-level visual cues extracted from images in guiding a controllable diffusion model for real-world image restoration and we design a multi-head control network adaptable via a mixture-of-experts strategy. We train our model without any prior assumption of specific degradations, through a smartly designed curriculum learning recipe. Additionally, we also introduce MetaRestore, a metalens imaging benchmark containing images with multiple degradations and artifacts. Extensive evaluations on several challenging datasets, including our benchmark, demonstrate that our method achieves significant performance gains and can robustly restore images with severe degradations. Project page: https://codejaeger.github.io/unicorn-gh

Figures (5)

• Figure 1: We present UniCoRN, a unified image restoration approach for handling multiple degradations simultaneously. Our approach shows robust performance on existing benchmarks and excels on MetaRestore, exhibiting multiple degradations. UniCoRN, based on stable diffusion, is corruption-agnostic, flexible, and scalable.
• Figure 2: Applying a sequential combination of single degradation image restoration models still struggle to faithfully recover an image corrupted by multiple degradations. UniCoRN, on the other hand, can recover from multiple degradations, without prior knowledge of the corruption type.
• Figure 3: Overview of our proposed UniCoRN model for unified multi-degradation image restoration.
• Figure 4: Qualitative results on the proposed MetaRestore metalens imaging benchmark. The results shown here are zero-shot; none of the models have been fine-tuned on the dataset. Best viewed if zoomed in.
• Figure 5: Qualitative results on single-degradation restoration tasks. Best viewed if zoomed in.