This repository provides a simple Pytorch implementation of plug and play ADMM with examples.

The notebook pnp_admm_example.ipynb contains a example in which a pretrained convnet gaussian denoiser is downloaded (120MB) and plugged into ADMM for solving a motion deblur, inpainting, and super-resolution problem. You can also view in Google Colab.

The notebook denoiser_training.ipynb contains code for training a denoiser from scratch using a subset of ImageNet as trainingset. The trainingset will be download automatically (250MB). You can also view in Google Colab.

	Degraded	PnP output	Ground truth
Motion deblur
Inpainting
Super-resolution

References:
S. Venkatakrishnan, C. Bouman, and B. Wohlberg, “Plug-and-play priors for model based reconstruction,” in Proc. IEEE Global Conference on Signal and Information Processing, 2013, pp. 945–948.