The dataset is available [here] (https://github.com/zalandoresearch/fashion-mnist).
Put the data in the directory modelname/data/fashion-mnist, modelname = ['VAE','DCGAN','WGAN','WGAN-GP']
- train.py -- train each model.
- GD_regular.py/GD_surfing.py -- apply surfing and regular gradient descent to solve
$f(x) = |G(x)-G(x_*)|^2$ . - compress.py -- apply surfing and regular gradient descent to solve
$f(x) = |AG(x)-Ay|^2$ , where$y$ is from test data. - compressX.py -- apply surfing and regular gradient descent to solve
$f(x) = |AG(x)-AG(x_*)|^2$ . - make-data.py -- generate data used to plot the surfaces in introduction.
The following codes are from https://github.com/hwalsuklee/tensorflow-generative-model-collections
- VAE.py
- GAN.py
- WGAN.py
- WGAN-GP.py
- ops.py
- prior_factory.py
- utils.py