To use SliceGAN open run_slicegan and edit information as requested. SliceGAN requires a single 2D training image of an isotropic microstructure, or three 2D images taken at perpendicular angles of an anisotropic microstructure. Images can be colour, grayscale or n-phase.
Use 'python run_slicegan 1' to train a new generator or 'python run_slicegan 0' to generate and save an example .tif file
If you are interested in trying new architectures, see the networks.py file
To adjust the training parameters or algorithm, see train.py
To add a new preprocessing method e.g for a different training datatype, see preproccessing.py.
Have to cd to a different repo that has a requirements file.
>> cd "C:\Users\pgasper\Documents\GitHub\SliceGAN\SliceGAN_AdaIN - REUSABILITY REPORT\SliceGAN_AdaIN-master"
Below is copied from their install shell script.
>> conda create -n sliceGAN python=3.7
>> conda activate sliceGAN
>> conda install pytorch==1.7.1 torchvision==0.8.2 torchaudio==0.7.2 cudatoolkit=10.2 -c pytorch
>> conda install -c anaconda ipykernel
>> python -m ipykernel install --user --name=sliceGAN
>> pip install -r requirements.txt
The conda installation of pillow library caused an error.
>> conda uninstall pillow
>> pip install pillow
v1.1 release:
v1.0 release:
