UnSCAR: Universal, Scalable, Controllable, and Adaptable Image Restoration
This work addresses the scalability and generalization limitations of universal image restoration models, which is a significant problem for researchers and practitioners aiming to deploy single models for diverse real-world image degradations.
This paper tackles the problem of scaling universal image restoration models to multiple degradations, where existing methods suffer from unstable training, large models, and performance drops. The authors propose a multi-branch mixture-of-experts architecture that decomposes restoration knowledge, enabling scalable learning over sixteen degradations and robust generalization to unseen domains.
Universal image restoration aims to recover clean images from arbitrary real-world degradations using a single inference model. Despite significant progress, existing all-in-one restoration networks do not scale to multiple degradations. As the number of degradations increases, training becomes unstable, models grow excessively large, and performance drops across both seen and unseen domains. In this work, we show that scaling universal restoration is fundamentally limited by interference across degradations during joint learning, leading to catastrophic task forgetting. To address this challenge, we introduce a unified inference pipeline with a multi-branch mixture-of-experts architecture that decomposes restoration knowledge across specialized task-adaptable experts. Our approach enables scalable learning (over sixteen degradations), adapts and generalizes robustly to unseen domains, and supports user-controllable restoration across degradations. Beyond achieving superior performance across benchmarks, this work establishes a new design paradigm for scalable and controllable universal image restoration.