Multi-scale feature fusion distilled attention network for efficient image super-resolution

Yinggan Tang, Mengjie Su, Xiuli Zhang

Abstract: Highlights•We proposed a multi-scale convolution (MConv) for learning multi-scale information.•We constructed a multi-scale feature fusion distillation block (MFFDB) using MConv.•We designed a distilled spatial attention (DSA) for spatial feature interactions.•We proposed a depth-wise channel attention (DCA) using a depth-wise convolution.•A large number of experiments show that the proposed MFFDAN has superior performance.