Projected WGAN-GP training

Pytorch implementation of WGAN-GP with a projection operator.

Acknowledgements

Python >= 3.6
Pytorch v1.0.0
Numpy
SciPy
tensorboardX (installation here). It is very convenient to see costs and results during training with TensorboardX for Pytorch
TensorFlow for tensorboardX
Use requirements.txt to install all requirements with pip or conda

gan_train.py: This model is mainly based on GoodGenerator and GoodDiscriminator of gan_64x64.py model from Improved Training of Wasserstein GANs. We modify this model for polycrystalline generation by adding a statistical projection loss and softmax activation to the generator.

Results such as costs, generated images (every 200 iters) for tensorboard will be written to ./runs folder.

To display the results to tensorboard, run: tensorboard --logdir runs

Name		Name	Last commit message	Last commit date
Latest commit History 32 Commits
.idea		.idea
DiFi @ b570669		DiFi @ b570669
libs		libs
models		models
.gitignore		.gitignore
.gitmodules		.gitmodules
LICENSE		LICENSE
README.md		README.md
Voronoi_p2_testing.py		Voronoi_p2_testing.py
_config.yml		_config.yml
baseline_gan.py		baseline_gan.py
data_analysis.ipynb		data_analysis.ipynb
gan_train.py		gan_train.py
gentest.py		gentest.py
gentest_p2.py		gentest_p2.py
matscidata.py		matscidata.py
p2_analysis.ipynb		p2_analysis.ipynb
p2_curves.h5		p2_curves.h5
polycrystaldata.py		polycrystaldata.py
requirements.txt		requirements.txt
results_analysis.ipynb		results_analysis.ipynb
tf_invariantdisc.py		tf_invariantdisc.py
utils.py		utils.py
voronoidata.py		voronoidata.py