Projected WGAN-GP training

Pytorch implementation of WGAN-GP with a projection operator.

Acknowledgements

• Modification of link
• igul222/improved_wgan_training
• caogang/wgan-gp
• LayerNorm

Prerequisites

• 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

Model

• 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.

Additional notes.

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