fagan2888 / geometricgan

Geometric GAN: first paper to introduce the hinge loss and geometric view of GAN

Geometric GAN

Code accompanying the paper "Geometric GAN".
(Ths code is modified from https://github.com/martinarjovsky/WassersteinGAN)

Prerequisites
Datasets
Reproducing Experiments
Generated Samples
Plot Losses

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.

Datasets

MNIST

Make empty folder at <PATH>/<TO>/<MNIST>.

Set symbolic link as follows;

mkdir data
ln -s <PATH>/<TO>/<MNIST> data/mnist

Note: you can leave the folder empty since torchvision will automatically download mnist dataset.

CelebA

Download Align&Cropped Images of CelebA dataset, i.e. img_align_celeba.zip, from https://drive.google.com/drive/folders/0B7EVK8r0v71pTUZsaXdaSnZBZzg at <PATH>/<TO>/<CelebA>.

unzip img_align_celeba.zip

Then you have,

<PATH>/<TO>/<CelebA>
├── img_align_celeba.zip
└── img_align_celeba

Set symbolic link as follows;

mkdir data
ln -s <PATH>/<TO>/<CelebA> data/celeba

LSUN

Download LSUN bedroom dataset using https://github.com/fyu/lsun at <PATH>/<TO>/<LSUN>.

unzip bedroom_train_lmdb.zip

Then you have,

<PATH>/<TO>/<LSUN>
├── bedroom_train_lmdb.zip
├── bedroom_train_lmdb
...

Set symbolic link as follows;

mkdir data
ln -s <PATH>/<TO>/<LSUN> data/lsun

Reproducing Experiments

Exp1: Mixture of Gaussian

python main.py standard geogan --cuda --dataset toy4 --dataroot '' --lrD 0.001 --lrG 0.001 --nc 2 --nz 4 --ngf 128 --ndf 128 --model_G toy4 --model_D toy4 --batchSize 500 --experiment samples/toy4_geogan_toy4_rmsprop_lr001_c1 --niter 500 --ndisplay 100 --nsave 50

or execute following scripts in the directory of this repo.

./scripts/exp1a.toy.all.sh
./scripts/exp1b.toy.diffC.sh

Exp2: MNIST

python main.py standard geogan --cuda --dataset mnist --dataroot data/mnist --imageSize 64 --nc 1 --lrD 0.0002 --lrG 0.0002 --model_G dcgan --model_D dcgan --ndf 128 --ngf 128 --Giters 10 --niter 25 --ndisplay 100 --nsave 5 --experiment samples/mnist_geogan_dcgan128_rmsprop_lr0002_kg10_c1

or execute following scripts in the directory of this repo.

./scripts/exp2.mnist.sh

Exp3: CelebA

python main.py standard geogan --cuda --dataset folder --dataroot data/celeba --loadSize 96 --imageSize 64 --lrD 0.0002 --lrG 0.0002 --model_G dcgan --model_D dcgan --ndf 128 --ngf 128 --Giters 10 --niter 50 --ndisplay 500 --nsave 5 --experiment samples/celeba_geogan_dcgan128_rmsprop_lr0002_kg10_c1

or execute following scripts in the directory of this repo.

./scripts/exp3.celeba.sh

Exp4: LSUN

python main.py standard geogan --cuda --dataset lsun --dataroot data/lsun --imageSize 64 --lrD 0.0002 --lrG 0.0002 --model_G dcgan --model_D dcgan --ndf 128 --ngf 128 --Giters 10 --niter 5 --nsave 1 --ndisplay 500 --experiment samples/lsun_geogan_dcgan128_rmsprop_lr0002_kg10_c1

or execute following scripts in the directory of this repo.

./scripts/exp4.lsun.sh

Generated Samples

Generated samples will be in the samples folder.

Plot Losses

Logs will be in the logs folder (if you use the aforementioned scripts).

Use plot_log.py, and the example usages of it are in scripts/plot.example.sh