Aharon Shai ; Ben-Artzi Gil

Abstract

Adaptive image restoration models can restore images with differentdegradation levels at inference time without the need to retrain the model. Wepresent an approach that is highly accurate and allows a significant reductionin the number of parameters. In contrast to existing methods, our approach canrestore images using a single fixed-size model, regardless of the number ofdegradation levels. On popular datasets, our approach yields state-of-the-artresults in terms of size and accuracy for a variety of image restoration tasks,including denoising, deJPEG, and super-resolution.