Implementation of our regularization method for training the StyleGAN2 model under the limited dataset.
Clone this repository
git clone https://github.com/google/lecam-gan.git
cd lecam-gan/stylegan2
Install packages
conda create --name lcgan_tf python=3.6
conda activate lcgan_tf
pip install -r requirements.txt
Copy the file lecam_loss.py
to the stylegan2 repository
Please refer to the instruction here for processing the dataset.
Specifically, please use to the num_samples
configuration in the
dataset_tool.py
file to build limited CIFAR dataset.
Please refer to the scripts we provide in the scripts
folder.
This repository is built based on the implementation from StyleGAN2-ADA.