This project consisted in studying GANs in the case of Wasserstein distance, as part of the fifth-year course at INSA Toulouse in Applied Mathematics of High Dimensional and Deep Learning.

GAN python scholarship

Go to file

Soumith Chintala 42fad7dcbb add readme		2017-01-30 20:11:11 +05:30
imgs	add readme	2017-01-30 20:11:11 +05:30
models	push code	2017-01-30 06:19:57 -08:00
LICENSE.md	add license	2017-01-30 19:51:35 +05:30
README.md	add readme	2017-01-30 20:11:11 +05:30
main.py	push code	2017-01-30 06:19:57 -08:00
requirements.txt	push code	2017-01-30 06:19:57 -08:00

README.md

Wasserstein GAN

Code accompanying the paper "Wasserstein GAN"

##Prerequisites

Computer with Linux or OSX
PyTorch
For training, an NVIDIA GPU is strongly recommended for speed. CPU is supported but training is very slow.

Two main empirical claims:

###Generator sample quality correlates with discriminator loss

###Improved model stability

##Reproducing LSUN experiments

With DCGAN:

python main.py --dataset lsun --dataroot [lsun-train-folder] --cuda

With MLP:

python main.py --mlp_G --ngf 512

Generated samples will be in the samples folder.

If you plot the value -Loss_D, then you can reproduce the curves from the paper. The curves from the paper (as mentioned in the paper) have a median filter applied to them:

med_filtered_loss = scipy.signal.medfilt(-Loss_D, dtype='float64'), 101)