DuNets-RMA

Code for the paper Deep Unrolling Networks with Recurrent Momentum Acceleration for Nonlinear Inverse Problems.

Requirements

In order to run the code, you will need the following:

• PyTorch (>= 1.10.0)
• Python (>=3.7.xx)
• PyEIT (link)
• pytorch_lightning

Structure

A nonlinear deconvolution problem

1. simulate data: toy_dataset_simulation.py
2. train: LPD-RMA toy_train_lpd.py; LPGD-RMA toy_train_lpgd.py

Electrical impedance tomography

1. simulate data: eit_circle_simulation.py
2. train: LPD-RMA eit_train_lpd.py; LPGD-RMA eit_train_lpgd.py
3. compute the quantitative metrics: eit_reconstruct_2inclusion.py and eit_reconstruct_4inclusion.py