Discriminative Activation of Information Is What You Need in Image Super-Resolution Transformer

Despite the significant progress made by Transformer in image super-resolution tasks, it has not effectively utilized prior knowledge in the image frequency domain and differentiated the processing of high-frequency and low-frequency information in the image. Previous studies on image super-resolution have shown that the high-frequency and low-frequency regions of the image exhibit distinct differences during the super-resolution process. In this paper, we propose a Discriminative Information Activation Super-Resolution Transformer (DIAST) to further improve the performance of Transformer in SISR tasks by discriminating high-frequency information from low-frequency information in images and discriminating cross-window information from inside-window information efficiently. Our results demonstrate that our method can further utilize the potential of the Transformer. The codes will be available at https://github.com/qyx1999/DIAST .