Generation of anime faces using WGAN-GP.

Try it yourself

• This project implements a Wasserstein GAN with Gradient Penalty (WGAN-GP) to generate 64 x 64 anime face images. This model uses a custom Generator and Critic architecture with PixelNorm.
• The model is trained on anime faces dataset from Kaggle.

Anime-Faces-GAN/
├── app.py                  # Front end
├── model.py                # Models' definitions
├── Models/                 # Models' save files
├── Model_Graphs/           # Models' graphs
├── Example_Results/        # Example images
├── trainer_kaggle.ipynb    # Trainer file
├── requirements.txt        # Dependencies
└── README.md               # This file

• app.py : Handles front end and deployment using streamlit python library.
• model.py : Contains Generator & Critic architecture's definitions.
• Models/ : Contains Generator's & Critic's save files from training.
• Model_Graphs/ : Contains Generator's & Critic's architecture graphs in svg.
• Example_Results/ : Contains a few example images (generated).
• trainer_kaggle.ipynb : Trainer file that runs on kaggle and generates required save files.

Click to expand the full Generator graph

Click to expand the full Critic graph

A few generated images.