Complexity Experts are Task-Discriminative Learners for Any Image Restoration

TL;DR

This paper introduces complexity experts in image restoration, enabling task-specific expert allocation with varying computational complexity, leading to improved performance and efficiency over existing methods.

Contribution

It proposes a novel flexible MoE architecture with complexity experts that adaptively assign tasks based on degradation complexity, enhancing generalization and inference efficiency.

Findings

Outperforms state-of-the-art methods in image restoration.

Enables effective bypassing of irrelevant experts during inference.

Demonstrates task-specific expert allocation driven by complexity bias.

Abstract

Recent advancements in all-in-one image restoration models have revolutionized the ability to address diverse degradations through a unified framework. However, parameters tied to specific tasks often remain inactive for other tasks, making mixture-of-experts (MoE) architectures a natural extension. Despite this, MoEs often show inconsistent behavior, with some experts unexpectedly generalizing across tasks while others struggle within their intended scope. This hinders leveraging MoEs' computational benefits by bypassing irrelevant experts during inference. We attribute this undesired behavior to the uniform and rigid architecture of traditional MoEs. To address this, we introduce ``complexity experts" -- flexible expert blocks with varying computational complexity and receptive fields. A key challenge is assigning tasks to each expert, as degradation complexity is unknown in advance. Thus, we execute tasks with a simple bias toward lower complexity. To our surprise, this preference effectively drives task-specific allocation, assigning tasks to experts with the appropriate complexity. Extensive experiments validate our approach, demonstrating the ability to bypass irrelevant experts during inference while maintaining superior performance. The proposed MoCE-IR model outperforms state-of-the-art methods, affirming its efficiency and practical applicability. The source code and models are publicly available at \href{https://eduardzamfir.github.io/moceir/}{\texttt{eduardzamfir.github.io/MoCE-IR/}}