Make model training reproducible

[valosekj]

Hey @KaterinaKrejci231054,

It would be great to make your multi-contrast model training as reproducible as possible. For example, you could create a download_data.py script to download specific dataset versions (commit ID or branch) and a shell script for nnUNet training.

Examples:

• https://github.com/ivadomed/model_seg_sci/blob/jv/enrich_SCIsegV2_with_praxis/dataset-conversion/download_data.py
• https://github.com/ivadomed/model_seg_sci/blob/jv/enrich_SCIsegV2_with_praxis/training_scripts/run_scisegv2.sh