Generative Models in NNabla

Collection of generative models in NNabla.

What's in it?

NNabla & Jupyter Notebook implementation of ...

Generative Adversarial Networks

1. Original GAN [notebook] [paper] (2014.06)
2. Conditional GAN [notebook] [paper] (2014.11)
3. Adversarially Learned Inference [notebook] [paper] (2016.06)
4. f-GAN [notebook] [paper] (2016.06)
5. InfoGAN [notebook] [paper] (2016.06)
6. Coupled GAN [notebook] [paper] (2016.06)
7. Energy Based GAN [notebook] [paper] (2016.09)
8. Auxiliary Classifier GAN [notebook] [paper] (2016.10)
9. Least Squares GAN [notebook] [paper] (2016.11)
10. Mode Regularized GAN [notebook] [paper] (2016.12)
11. Generative Adversarial Parallelization [notebook] [paper] (2016.12)
12. Wasserstein GAN [notebook] [paper] (2017.01)
13. Boundary Seeking GAN [notebook] [paper] (2017.02)
14. DiscoGAN [notebook] [paper] (2017.03)
15. Boundary Equilibrium GAN [notebook] [paper] (2017.03)
16. DualGAN [notebook] [paper] (2017.04)
17. Margin Adaptation for GAN [notebook] [paper] (2017.04)
18. Softmax GAN [notebook] [paper] (2017.04)

Variational AutoEncoders

1. Original VAE [notebook] [paper] (2013.12)
2. Conditional VAE [notebook] [paper] (2014.06)
3. Adversarial Autoencoder [notebook] [paper] (2015.11)
4. Denoising VAE [notebook] [paper] (2015.11)
5. Adversarial Variational Bayes [notebook] [paper] (2017.01)

VAE-GANs

1. α-GAN [notebook] [paper] (2017.06)

Dependencies

• jupyter == '1.0.0'
• nnabla == '0.9.1rc3'
• scipy == '0.19.1'
• numpy == '1.13.0'
• tensorflow == '1.2.0' (for MNIST data only)

TODO

• parameter search
• weight clipping
• Improved Training of Wasserstein GANs (2017.04)

Reference

This repository is inspired by wiseodd's awesome implementation in Pytorch & Tensorflow