Improved differentiable neural architecture search for single image super-resolution

Abstract Deep learning has shown prominent superiority over other machine algorithms in Single Image Super-Resolution (SISR). In order to reduce the efforts and resources cost on manually designing deep architecture, we use differentiable neural architecture search (DARTS) SISR. Since was originally used for classification tasks, our experiments show that direct usage of DARTS super-resolutions tasks will give rise many skip connections which results poor performance final architecture. Thus, it is necessary have made some improvements application field According characteristics SISR, remove redundant operations redesign cell achieve an improved DARTS. Then convolution cells as a nonlinear mapping part super-resolution network. The new shows its effectiveness benchmark datasets DIV2K dataset.