Deeply-Recursive Convolutional Network for Image Super-Resolution

TL;DR

This paper introduces a deeply-recursive convolutional network for image super-resolution that leverages high recursion depth and novel training techniques to significantly outperform previous methods.

Contribution

It presents a new recursive network architecture with recursive-supervision and skip-connection to improve training and performance in image super-resolution.

Findings

Outperforms previous super-resolution methods by a large margin

Deep recursion improves performance without extra parameters

Recursive-supervision and skip-connection ease training difficulties

Abstract

We propose an image super-resolution method (SR) using a deeply-recursive convolutional network (DRCN). Our network has a very deep recursive layer (up to 16 recursions). Increasing recursion depth can improve performance without introducing new parameters for additional convolutions. Albeit advantages, learning a DRCN is very hard with a standard gradient descent method due to exploding/vanishing gradients. To ease the difficulty of training, we propose two extensions: recursive-supervision and skip-connection. Our method outperforms previous methods by a large margin.