Recent works that utilized deep models have achieved superior results in various image restoration (IR) applications. Such approach is typically supervised, which requires a corpus of training images with distributions similar to the images to be recovered. On the other hand, the shallow methods, which are usually unsupervised remain promising performance in many inverse problems, \eg, image deblurring and image compressive sensing (CS), as they can effectively leverage nonlocal self-similarity priors of natural images.