Skip to content

[ICML 2019] Plug-and-Play Methods Provably Converge with Properly Trained Denoisers

Notifications You must be signed in to change notification settings

uclaopt/Provable_Plug_and_Play

Repository files navigation

Provable_Plug_and_Play

The implement of the following paper:

E. K. Ryu, J. Liu, S. Wang, X. Chen, Z. Wang, and W. Yin. "Plug-and-Play Methods Provably Converge with Properly Trained Denoisers." ICML, 2019.

Scripts

  1. pnp_admm_csmri.py (CS-MRI solved with Plug-and-Play ADMM)
  2. pnp_fbs_csmri.py (CS-MRI solved with Plug-and-Play FBS)
  3. pnp_admm_poisson_denoise.py (Poisson Denoising solved with Plug-and-Play ADMM)
  4. pnp_fbs_poisson_denoise.py (Poisson Denoising solved with Plug-and-Play FBS)
  5. pnp_admm_photon_imaging.py (Single Photon Imaging solved with Plug-and-Play ADMM)
  6. pnp-fbs_photon_imaging.py (to appear soon)
  7. training/train_full_realsn.py (Training the denoisers)

How to run the scripts

Run with default settings

$ python3 pnp_admm_csmri.py

Run with costmized settings

$ python3 pnp_admm_csmri.py --model_type DnCNN --sigma 15 --alpha 2.0 --maxitr 100 --verbose 1

All the arguments are explained in the file "utils/config.py".

Training

We provide some pretraining models in the folder "Pretrained_models". They can be directly used in the Plug-and-PLay framework. To train new models, please refer the "README" file in the "training" folder.

Citation

If you find our code helpful in your resarch or work, please cite our paper.

@InProceedings{pmlr-v97-ryu19a,
  title = 	 {Plug-and-Play Methods Provably Converge with Properly Trained Denoisers},
  author = 	 {Ryu, Ernest and Liu, Jialin and Wang, Sicheng and Chen, Xiaohan and Wang, Zhangyang and Yin, Wotao},
  booktitle = 	 {Proceedings of the 36th International Conference on Machine Learning},
  pages = 	 {5546--5557},
  year = 	 {2019},
  editor = 	 {Chaudhuri, Kamalika and Salakhutdinov, Ruslan},
  volume = 	 {97},
  series = 	 {Proceedings of Machine Learning Research},
  address = 	 {Long Beach, California, USA},
  month = 	 {09--15 Jun},
  publisher = 	 {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v97/ryu19a/ryu19a.pdf},
  url = 	 {http://proceedings.mlr.press/v97/ryu19a.html}
}

About

[ICML 2019] Plug-and-Play Methods Provably Converge with Properly Trained Denoisers

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published