/generative.models.tensorflow.v2

Primary LanguageJupyter NotebookApache License 2.0Apache-2.0

Generative models with tensorflow version 2.0 style

  • Final update: 2019. 09. 26.
  • All right reserved @ Il Gu Yi 2019

This repository is a collection of various generative models (GAN, VAE, Normalizing flow, Autoregressive models, etc) implemented by TensorFlow version 2.0 style

Getting Started

Prerequisites

  • TensorFlow 2.0.0-rc1 (except normalizing_flow/nice.ipynb which is based on tf version 1.13.1)
  • Python 3.6
  • Python libraries:
    • numpy, matplotlib, PIL, imageio
    • urllib, zipfile
  • TensorFlow libraries & extensions:
  • Jupyter notebook
  • OS X and Linux (Not validated on Windows OS)

Contents

Generative Adversarial Networks (GANs) [with MNIST and Fashion MNIST]

GAN

MNIST Fashion MNIST

DCGAN (Deep Convolutional GAN)

MNIST Fashion MNIST

Conditional GAN

MNIST
Fashion MNIST

LSGAN

MNIST Fashion MNIST

BiGAN

MNIST
Fashion MNIST

Wasserstein GAN

MNIST Fashion MNIST

WGAN-GP

MNIST Fashion MNIST

Pix2Pix (Image Translation)

facades
cityspaces

CycleGAN (Unpaired Image Translation)

Latent Variable Models [with MNIST]

AutoEncoder (actually not generative model)

MNIST
Fashion MNIST

Denosing AutoEncoder

AutoRegressive Models [with MNIST]

Fully Visible Sigmoid Belief Networks

MNIST Fashion MNIST

Neural Autoregressive Density Estimation

MNIST Fashion MNIST

Normalizing Flow Models [with MNIST]

NICE: Non-Linear Independent Components Estimation

Author

Il Gu Yi

Slides

Notion link