ShuffleMixer: An Efficient ConvNet for Image Super-Resolution

TL;DR

ShuffleMixer is a lightweight, efficient convolutional neural network for image super-resolution that uses large depth-wise convolutions combined with channel split-shuffle operations, achieving high performance with significantly fewer parameters.

Contribution

It introduces a novel large kernel ConvNet architecture with channel split-shuffle and Fused-MBConvs for mobile-friendly image super-resolution, outperforming existing models in efficiency.

Findings

6x fewer parameters than state-of-the-art methods

Achieved competitive super-resolution performance

Won the NTIRE 2022 model complexity track

Abstract

Lightweight and efficiency are critical drivers for the practical application of image super-resolution (SR) algorithms. We propose a simple and effective approach, ShuffleMixer, for lightweight image super-resolution that explores large convolution and channel split-shuffle operation. In contrast to previous SR models that simply stack multiple small kernel convolutions or complex operators to learn representations, we explore a large kernel ConvNet for mobile-friendly SR design. Specifically, we develop a large depth-wise convolution and two projection layers based on channel splitting and shuffling as the basic component to mix features efficiently. Since the contexts of natural images are strongly locally correlated, using large depth-wise convolutions only is insufficient to reconstruct fine details. To overcome this problem while maintaining the efficiency of the proposed module, we introduce Fused-MBConvs into the proposed network to model the local connectivity of different features. Experimental results demonstrate that the proposed ShuffleMixer is about 6x smaller than the state-of-the-art methods in terms of model parameters and FLOPs while achieving competitive performance. In NTIRE 2022, our primary method won the model complexity track of the Efficient Super-Resolution Challenge [23]. The code is available at https://github.com/sunny2109/MobileSR-NTIRE2022.