Crafting a Toolchain for Image Restoration by Deep Reinforcement Learning

TL;DR

This paper introduces RL-Restore, a reinforcement learning-based method that dynamically selects from a toolbox of small networks to progressively restore corrupted images, outperforming traditional fixed models especially on complex distortions.

Contribution

It proposes a novel reinforcement learning framework with a toolbox of specialized networks and a joint training scheme for adaptive, efficient image restoration.

Findings

RL-Restore effectively handles complex, unknown distortions.

It achieves better parameter efficiency than conventional networks.

The method demonstrates improved restoration quality in experiments.

Abstract

We investigate a novel approach for image restoration by reinforcement learning. Unlike existing studies that mostly train a single large network for a specialized task, we prepare a toolbox consisting of small-scale convolutional networks of different complexities and specialized in different tasks. Our method, RL-Restore, then learns a policy to select appropriate tools from the toolbox to progressively restore the quality of a corrupted image. We formulate a step-wise reward function proportional to how well the image is restored at each step to learn the action policy. We also devise a joint learning scheme to train the agent and tools for better performance in handling uncertainty. In comparison to conventional human-designed networks, RL-Restore is capable of restoring images corrupted with complex and unknown distortions in a more parameter-efficient manner using the dynamically formed toolchain.