InnerEyeCam

Package to simplify installation and use of InnerEye-DeepLearning applications. This package was originally written for MacOS 12.4, but has also been tested on CentOS 8.5.2111.

This package contains the following:

• install.sh: script for installing InnerEye-DeepLearning;
• ML: directory containing modules modified with respect to the InnerEye-DeepLearning originals, and containing a configs sub-directory for storing model configurations;
• environment.yml : file defining creation of 'conda' environment for running InnerEye-DeepLearning applications;
• runner.py : script for running InnerEye-DeepLearning applications;
• settings.py : settings for configuration variables used when running InnerEye-DeepLearning applications;
• conda-setup.sh : script for activating the conda environment for running InnerEye-DeepLearning applications;
• train.sh : script for executing a InnerEye-DeepLearning training run.
• test: directory for scripts to test installation.
• README.md : file containing this help information.

Installation

1. Clone the InnerEyeCam repository:

   git clone
   

2. Ensure that a working installation of conda is available. By default conda tools for MacOS are assumed to be in the directory: /opt/miniconda3/. If this isn't the case, then the file InnerEyeCam/conda-setup.sh needs to be changed to reflect the actual location.

3. Change to the directory InnerEyeCam:

   cd InnerEyeCam
   

4. Run the install script:

   ./install.sh
   

   This performs the following operations:

   • It clones InnerEye-DeepLearning into the same directory as InnerEyeCam, deleting any pre-existing clone.

   -If, in the script, CHECKOUT_VERSION is set to 1, then the version of the code specified by INNEREYE_VERSION is checked out.

   • It recursively copies the InnerEyeCam directory to `InnerEye-DeepLearning'.

   • Within InnerEye-DeepLearning, It copies from InnerEyeCam/v0.x/ML to InnerEye/ML the files:

     model_testing.py
     dataset/full_image_dataset.py
     visualizers/plot_cross_validation.py # version v0.3 only
     

     The first two have changes with respect to the InnerEye-DeepLearning originals, to enable multi-threading under MacOS. The third has changes to avoid crashes in cases of datasets with CSV_SERIES_HEADER and/or CSV_INSTITUTION_HEADER undefined.

   • It copies InnerEyeCam/v0.x/environment.yml and (v0.4 only) InnerEyeCam/v0.x/primary_deps_mac.yml to InnerEye-DeepLearning`.

   • It creates a conda environment for running InnerEye-DeepLearning applications.

5. After installation, the directory structure should be as follows:

                   |
        ----------------------
        |                    |       
   InnerEyeCam     InnerEye-DeepLearning
                             |
                      ----------------
                      |              |
                   InnerEye     InnerEyeCam
   

   When running on Azure, it's expected that all code should be in a single directory tree: InnerEye-DeepLearning. To fit in with this, subsequent user modifications should be in InnerEye-DeepLearning/InnerEyeCam.

Preparing to run InnerEye-DeepLearning

1. Add any model configurations to be used to the directory InnerEye-DeepLearning/InnerEyeCam/ML/configs/segmentation.

2. Edit as needed InnerEye-DeepLearning/InnerEyeCam/settings.yml. For explanation of settings, see:

   • How to setup Azure Machine Learning for InnerEye - Step 6

3. Edit as needed InnerEye-DeepLearning/InnerEyeCam/train.sh. This includes examples of commands for running InnerEye-DeepLearning applications locally and on Azure, with explanations of the parameters used.

Performing a training run

1. Edit InnerEye-DeepLearning/InnerEyeCam/runner.py, and ensure that the value of innereye_version matches the version of InnerEye-DeepLearning installed.

2. Execute the script InnerEye-DeepLearning/InnerEyeCam/train.sh.

3. If submitting to Azure, progress can be monitored at:

   • https://ml.azure.com/