This is the pytorch implementation of 3 different GAN models using same convolutional architecture.
- DCGAN (Deep convolutional GAN)
- WGAN-CP (Wasserstein GAN using weight clipping)
- WGAN-GP (Wasserstein GAN using gradient penalty)
The prominent packages are:
- numpy
- scikit-learn
- tensorflow 1.5.0
- pytorch 0.3.0
- torchvision 0.3.0
To install all the dependencies quickly and easily you should use pip
pip install -r requirements.txtRunning training of DCGAN model on Fashion-MNIST dataset:
python main.py --model DCGAN \
--is_train True \
--download True \
--dataroot datasets/fashion-mnist \
--dataset fashion-mnist \
--epochs 30 \
--cuda True \
--batch_size 64
Running training of WGAN-GP model on CIFAR-10 dataset:
python main.py --model WGAN-GP \
--is_train True \
--download True \
--dataroot datasets/cifar \
--dataset cifar \
--generator_iters 40000 \
--cuda True \
--batch_size 64
Start tensorboard:
tensorboard --logdir ./logs/
Interpolation between a two random latent vector z over 10 random points, shows that generated samples have smooth transitions.
