1-class and 2-class segmentation of axon/myelin using nnunetv2

First create a virtual environment (using pipenv, conda, etc.). Note that this project requires a python version >= 3.9 Then, install the requirements:

pip install -r requirements.txt

The inference tool should now be ready to use. First, download a model using the following command. The user will be prompted to specify which model to download.

python download_models.py

Then, you can use the nn_axondeepseg.py script to apply the model to your images. Assuming the images are in a folder called input, you can use

python nn_axondeepseg.py --seg-type UM --path-out output-folder --path-dataset input

The --seg-type argument is used to specify which kind of model is used: UM stands for unmyelinated axon, for which we expect a single class output; AM stands for axon and myelin, for which we expect a 2-class output. The user can specify any nnUNet model using the --path-model argument.