An pytorch implementation of Paper "Improved Training of Wasserstein GANs".

Python, NumPy, SciPy, Matplotlib A recent NVIDIA GPU

A latest master version of Pytorch

• gan_toy.py : Toy datasets (8 Gaussians, 25 Gaussians, Swiss Roll).(Finished in 2017.5.8)

• gan_language.py : Character-level language model (Discriminator is using nn.Conv1d. Generator is using nn.Conv1d. Finished in 2017.6.23. Finished in 2017.6.27.)

• gan_mnist.py : MNIST (Running Results while Finished in 2017.6.26. Discriminator is using nn.Conv1d. Generator is using nn.Conv1d.)

• gan_64x64.py: 64x64 architectures(Looking forward to your pull request)

• gan_cifar.py: CIFAR-10(Looking forward to your pull request)

• Toy Dataset

  Some Sample Result, you can refer to the results/toy/ folder for details.

  • 8gaussians 154500 iteration

  

  • 25gaussians 48500 iteration

    

  • swissroll 69400 iteration

  

• Mnist Dataset

  Some Sample Result, you can refer to the results/mnist/ folder for details.

  

  

  

  

• Billion Word Language Generation (Using CNN, character-level)

  Some Sample Result after 8699 epochs which is detailed in sample

  I haven't run enough epochs due to that this is very time-comsuming.

  He moved the mat all out clame t

  A fosts of shores forreuid he pe

  It whith Crouchy digcloued defor

  Pamreutol the rered in Car inson

  Nor op to the lecs ficomens o fe

  In is a " nored by of the ot can

  The onteon I dees this pirder ,

  It is Brobes aoracy of " medurn

  Rame he reaariod to thim atreast

  The stinl who herth of the not t

  The witl is f ont UAy Y nalence

  It a over , tose sho Leloch Cumm

Based on the implementation igul222/improved_wgan_training and martinarjovsky/WassersteinGAN