Deep learning through domain-transform manifold learning for image reconstruction (AUTOMAP) is unstable

Code related to the paper "Deep learning through domain-transform manifold learning for image reconstruction (AUTOMAP) is unstable".

Setup

The data related to the paper can be downloaded from here. After downloading the data, modify the paths in the file adv_tools_PNAS/automap_config.py so that all relevant paths points to the data. To run the stability test for the LASSO experiment, add the UiO-CS/optimization and UiO-CS packages to your Python path.

Overview of the different files

---

• Figure 1: Demo_test_automap_stability.py
• Figure 2: Demo_test_automap_random_noise.py and Demo_test_lasso_random_noise.py
• Figure 3: Demo_test_lasso_stability.py and Demo_test_lasso_on_automap_pert.py
• Figure 4: Demo_test_automap_random_noise.py and Demo_test_lasso_random_noise.py
• SI Table 2: Demo_test_automap_compute_norms.py

---

All scripts have been exectured with Tensorflow version 1.14.0.