Material for course on Deep Learning in Scientific Inverse Problems, to be taught at Summer school on Data Assimilation.
The main teaching material is available in the form of Jupiter slides. Simply type jupyter nbconvert Lecture.ipynb --to slides --post serve --embed-images to access the slides.
Several tutorials are presented during the course in the form of Jupyter notebooks.
| Session | Exercise (Github) | Exercise (Colab) |
|---|---|---|
| EX0: Prepare brain dataset | Link | |
| EX1: Prepare brain-fbp dataset | Link | |
| EX2: Variational CTscan imaging | Link |  |
| EX3: Supervised Learning for CTscan imaging | Link | |
| EX4: DIP CTscan imaging | Link | |
| EX5: PnP for CTscan imaging | Link | |
| EX6: Learned iterative solver for CTscan imaging | Link | |
If you are attending the course, we will provide you with a GDrive link with a minimal dataset used in the examples (this is produced by the Prepare_brain_dataset.ipynb and Prepare_brainfbp_dataset.ipynb notebooks.
If you would like to run the entire pipeline (including the Prepare_brain_dataset.ipynb and Prepare_brainfbp_dataset.ipynb), you will need access to the original FASTMRI dataset. We will be working with a small subset of it that you can retrieve following these two simple steps:
- Register at the bottom of the website to obtain a list of links to be used to retrieve the dataset of interest. You will receive an email with instructions on how to retrieve the dataset;
- Run the curl command for the
brain_multicoil_val_batch_0.tar.xzdataset (be prepared to wait long time, and ensure you have 94GB of space in your disk).
To run the different Jupyter notebooks, participants can either use:
- local Python installation (simply run
./install_env.sh). Note, this requires access to a GPU. For CPU-only workstation, modify theenvironment.ymlfile accordingly. - a Cloud-hosted environment such as Google Colab (use links provided above to open the notebook directly in Colab). Before getting started, make sure to manually upload all .py files from the
notebooksdirectory and the entiremodeldirectory into your Colab local storage. Moreover, place the folder with the data that you have previously downloaded in your personal GDrive.