Unsupervised Image Super-Resolution using Cycle-in-Cycle Generative Adversarial Networks

TL;DR

This paper introduces an unsupervised image super-resolution method using a novel Cycle-in-Cycle GAN architecture, capable of handling noisy, blurry low-resolution images without paired training data, achieving competitive results.

Contribution

It proposes a new unsupervised framework with a Cycle-in-Cycle GAN structure for super-resolution without paired data or kernel estimation.

Findings

Achieves comparable results to supervised models on NTIRE2018 datasets.

Handles noisy and blurry low-resolution images effectively.

Operates without requiring paired low/high-resolution training data.

Abstract

We consider the single image super-resolution problem in a more general case that the low-/high-resolution pairs and the down-sampling process are unavailable. Different from traditional super-resolution formulation, the low-resolution input is further degraded by noises and blurring. This complicated setting makes supervised learning and accurate kernel estimation impossible. To solve this problem, we resort to unsupervised learning without paired data, inspired by the recent successful image-to-image translation applications. With generative adversarial networks (GAN) as the basic component, we propose a Cycle-in-Cycle network structure to tackle the problem within three steps. First, the noisy and blurry input is mapped to a noise-free low-resolution space. Then the intermediate image is up-sampled with a pre-trained deep model. Finally, we fine-tune the two modules in an end-to-end manner to get the high-resolution output. Experiments on NTIRE2018 datasets demonstrate that the proposed unsupervised method achieves comparable results as the state-of-the-art supervised models.