Single image super-resolution (SISR) deals with a fundamental problem of a low-resolution (LR) image to its high-resolution (HR) version. few years have witnessed impressive progress propelled by deep learning. However, one critical challenge faced by existing methods is to strike sweet spot of deep model complexity and resulting SISR quality. This paper this pain point by proposing a linearly-assembled pixel-adaptive network (LAPAR), which casts the direct LR to HR mapping learning a linear coefficient regression task over a dictionary of multiple filter bases. Such a parametric representation renders our model lightweight and easy to optimize while achieving state-of-the-art on SISR benchmarks. Moreover, based on the same idea, LAPAR is extended tackle other restoration tasks, e.g., image denoising and JPEG image, and again, yields strong performance. The code is available at://github.com/dvlab-research/Simple-SR.
No takes yet. Share an insight, caveat, or question.
Li et al. (2021) studied this question.