soobinseo / GANs

A tensorflow implementation of GANs with variable loss function

GANs

Description

A tensorflow implementation of GANs with variable loss function including Standard GAN, Least-Squared GAN (LSGAN), Wasserstein GAN (WGAN), improved Wasserstein GAN, and DRAGAN

Datasets

I used celebA dataset and cropped the center face by 64X64 pixels.

Results

• Standard GAN

• LSGAN

• WGAN

• Improved WGAN

Dependencies

1. tensorflow >= 1.0.0
2. numpy >= 1.12.0

File description

• hyperparams.py includes all hyper parameters that are needed.
• data.py loads training data and crops them.
• modules.py contains customized conv net and so on.
• networks.py builds a generator and a discriminator.
• train.py is for training.

Training the network

• STEP 1. Adjust hyper parameters in hyperparams.py, especially the hyperparameter "loss" that you want to train.
• STEP 2. create 'data' directory, then download and extract celebA data at the directory.
• STEP 3. Run train.py and show result through tensorboard.