Anime-GAN

This is a simple PyTorch implementation of a DCGAN, modified from a project in the Udacity DLND. The model is trained using an Anime Faces dataset I found on Kaggle.

The architecture is simple, the discriminator and the generator each has 4 conv layers and 1 fully connected layer. This is a first attempt of creating a DCGAN. A lot of improvements can be made in the future.

Instructions

1. Clone this repository and navigate to the folder.

   git clone https://github.com/davidxuan/Anime-GAN.git

   cd Anime-GAN

2. Use train_gan.ipynb to train the models.

   jupyter notebook tran_gan.ipynb

3. Use generate_samples.ipynb to load the models and generate some faces.

   jupyter notebook generate_samples.ipynb

4. If you wish to change the GAN architecture, modify model.py.