Lightweight and Accurate Single Image Super-Resolution with Channel Segregation Network

Abstract

Deep neural networks have witnessed great success in Single Image Super-Resolution (SISR). However, current improvements are mainly contributed by much deeper networks, which leads to huge computation cost and limited application for mobile devices. Moreover, most existing methods propagate the basic content of low-resolution images forward to deeper layers iteratively. Such duplicate computations inevitably result in inefficient reconstruction. To address this issue, we propose an efficient channel segregation block containing multiple branches with different depths, enabling the model to preserve basic content, and focusing on optimizing the detail content with fewer parameters. By merging the output of segregated branches, the block covers a large range of receptive fields. Experimental results demonstrate that the proposed method outperforms other state-of-the-art methods even with fewer parameters and lower computational complexity, which is more applicable to lightweight scenarios.