This paper highlights that despite the advancements in deep learning in the field of Single Image Super-Resolution (SISR), existing research has focused solely on performance enhancement and neglected quantifying the transferability of modules. We introduce the concept of "universality" and its definition, extending the existing concept of "generalization" to the transferability of modules. We also propose the "universality evaluation equation (UAE)," a metric that quantifies the transferability of modules. Based on the UAE results, we design two optimized modules: the Cycle Residual Block (CRB) and the Depth-Specific Cycle Residual Block (DCRB). Experiments on natural scene benchmarks, remote sensing datasets, and other low-level tasks demonstrate that the network with the proposed plug-and-play module outperforms several state-of-the-art methods, achieving up to 0.83 dB of PSNR improvement or a 71.3% parameter reduction. By applying a similar optimization approach to various base modules, we propose a new paradigm for plug-and-play module design.
