Lightweight Image Super-Resolution With Mobile Share-Source Network

Abstract

Within the development of the deep convolutional neural network, the great achievements had been made in the single-image super-resolution (SISR) task. However, the higher performance always comes with the deeper layers which also brings larger numbers of network operations and parameters that make it hard to implement in practice. In our work, a lightly super-resolution, named Mobile Share-Source Network (MSSN), is purposed to address these practical issues. In MSSN, a high-efficiency block, the mobile adaptive weighted residual unit, is designed to fulfill the need for the reduction in both parameters and the Mult-Adds while maintaining the performance with importing the deep separable convolution. Moreover, it brings into the Adaptive Weighted Share-Source Skip Connection, getting abundant information from the shallow layer which helps reconstruct better images. The experimental results show that our network has fewer numbers of parameters and operations than the state-of-the-art lightweight network while maintaining high reconstruction quality comparing with many state-of-the-art super-resolution methods in terms of both peak signal-to-noise ratio (PSNR) and structural similarity index metrics (SSIM).

Block-Attentive Subpixel Prediction Networks for Computationally Efficient Image Restoration

“…If the channel is expanded by two, the overall complexity is increase by four times. Therefore, like previous studies [22], [58], the channel dimension can be treated as complexity-performance control hyper-parameter.…”

“…This method is closely related to our approach, but there is considerable room to improve the speed-performance trade-off. Although many studies have been proposed for lightweight image restoration [55]- [58], they have generally focused on image super-resolution and network-related parameter reducing, as in image recognition This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ This article has been accepted for publication in a future issue of this journal, but has not been fully edited.…”

Efficient Spatial Pyramid of Dilated Convolution and Bottleneck Network for Zero-Shot Super Resolution

“…These approaches rely on a large number of samples and a multi-layer structure to learn the parameters of the network, which is suitable for some tasks such as target detection, semantic recognition and pose estimation, where we want to obtain comprehensive features to get better performance. Recently, more lightweight networks [28] have been explored to reduce parameters and enlarge receptive field. Inception module [29] concatenates all scale features to fuse multi-scale information learned by dilated convolutions to increase the reception field of our network and catch more contextual information.…”