AdaGAN: greedy iterative procedure to train mixtures of GANs
Python
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
models
.gitignore
LICENSE
README.md
adagan.py
adagan_cifar.py
adagan_cj.py
adagan_gmm.py
adagan_icml.py
adagan_mnist.py
adagan_mnist3.py
celeba_pot.py
cifar_sota.py
datahandler.py
debug_models.py
fid_pics.py
gan.py
iclr_celeba_began.py
iclr_celeba_dcgan.py
iclr_celeba_vae.py
iclr_mnist_vae.py
metrics.py
ops.py
pot.py
utils.py
vae.py

README.md

AdaGAN: Boosting Generative Models

This project implements the AdaGAN algorithm, presented in this paper.

Getting started

It is assumed that the user runs TensorFlow of version > 12.

Make sure the directory where you run code also contains sub-directories called mnist and models containing MNIST datasets and the pre-trained MNIST classifier respectively (provided in this repo).

Running

Gaussian Mixture Experiments

python adagan_gmm.py [options]

MNIST Experiments

python adagan_mnist.py [options]

3-digit MNIST Experiments

python adagan_mnist3.py [options]

Code is still under development and can be modified