Aggregate-and-Excite Network for Image Super-Resolution

Recently, convolution neural network (CNN)-based image super-resolution (SR) has achieved excellent performance. However, most existing works have ignored the importance of accurately utilizing contextual information. To solve this problem, we propose Aggregate-and-Excite block (AEB) with integration into our proposed Aggregate-and-Excite Network (AENet). AEB is functionally equivalent to feature extraction block. Each contains a global contextual information unit (GCIU) and a gate unit mechanism (GU), capturing the enriched contextual information for more powerful feature representations and emphasizing informative features according to the global contextual information. Compared with other methods, our AENet has achieved good performance through extensive experiments.