Deep Convolutional GANs with PyTorch

Description

Pytorch implementation of the deep convolutional GAN paper titled "Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks" using PyTorch. Link to the original paper by Radford, A., Metz, L. and Chintala, S. You can also find the vanilla GAN paper by Goodfellow, I. et al. for training details.

How to run

First, install dependencies (a new python virtual environment is recommended).

# clone project   
git clone https://github.com/visualCalculus/deep-convolutional-gan

# install project   
cd deep-convolutional-gan
pip install -e .
pip install -r requirements.txt

Next, navigate to dcgan folder and run train.py with appropriate command line arguments.

# module folder
cd dcgan

# run module
python train.py --gpus=1 --max_epochs=5 --dataset=MNIST --image_size=32 --batch_size=128 --z_dim=100
--z_filter_shape=2 --n_channels=1 --learning_rate=2e-4 --beta1=0.5

Results

DCGAN trained for 10 epochs on MNIST.

DCGAN trained for 50 epochs on CIFAR10.

Citation

@misc{visualCalculus2021dcgan,
  title={deep-convolutional-gan},
  author={Tejomay, Abhiroop},
  publisher={GitHub}
  journal={GitHub repository},
  year={2021},
  howpublished={\url{https://github.com/visualCalculus/deep-convolutional-gan}}
}