We provide the codes necessary to reproduce part of the experiments detailed in our EUSIPCO tutorial "Proximal Neural Networks: Wedding Variational Methods and Artificial Intelligence" and additional codes to deepen the understanding. Utilizing the DeepInverse library enables us to concentrate on the structure of iterative schemes while leveraging established knowledge about data-term and prior design, as well as their associated gradient and proximity operators.
Proximal algorithms
-
Forward-backward
- Example in image restoration (Blur + Gaussian noise) with TV-L12 denoiser. [Code Python] [Notebook Google Colab] -
FISTA
- Example in image restoration (Blur + Gaussian noise) with TV-L12 denoiser. [Code Python] [Notebook Google Colab] -
Douglas-Rachford
- Example in image restoration (Blur + Gaussian noise) with TV-L12 denoiser. [Code Python] -
Loris-Verhoeven
- Example in image restoration (Blur + Gaussian noise) with TV-L12 denoiser. [Code Python] -
Condat-Vu
- Example in image restoration (Blur + Gaussian noise) with TV-L1 denoiser. [Code Python]
- Example in image restoration (Blur + Gaussian noise) with TV-L12 denoiser. [Code Python] [Notebook Google Colab] -
Chambolle-Pock
- Example in image restoration (Blur + Gaussian noise) with TV-L12 denoiser. [Code Python]
Plug-and-play
-
Forward-backward PnP
- Example in image restoration (Blur + Gaussian noise) with DRUnet denoiser. [Code Python] [Notebook Google Colab]
- Example in image restoration (Blur + Gaussian noise) with GSDRUNet denoiser. [Code Python]
- Example in image restoration (Blur + Gaussian noise) with DnCNN denoiser. [Code Python] [Notebook Google Colab]
Unfolded
-
Example in image restoration (Blur + Gaussian noise).
- Unfolded CV for regularisation parameter estimation. [Code Python] [Notebook Google Colab]
Miscellaneous
-
Read different types of images with DeepInverse
- Examples for .html, .png... files [Code Python]
-
Create a forward model
- Inpainting [Code Python] [Notebook Google Colab]
- Blur [Code Python] [Notebook Google Colab]
- Blur and Inpainting [Code Python] [Notebook Google Colab]
- Tomography [Code Python] [Notebook Google Colab]
Contacts
-
Audrey Repetti
A.Repetti@hw.ac.uk
Heriot-Watt University & Maxwell Institute for Mathematical Sciences, Edinburgh, UK -
Nelly Pustelnik
nelly.pustelnik@ens-lyon.fr
CNRS, Laboratoire de Physique ENS Lyon -
Jean-Christophe Pesquet
jcpesquetpro@gmail.com
CentraleSupélec, Université Paris-Saclay, Gif-sur-Yvettes, France