Repository with code to reproduce the results for SMUG in our paper and our upcoming journal version
In this work, we propose SMUG that systematically integrates RS with MoDL using an deep unrolled architecture. We study in detail where to apply RS in the unrolled architecture for better performance and propose a novel unrolling loss to improve training efficiency. We show that the purposed SMUG is significantly effective in improving three major types of instabilities of MoDL.