The search code will be published once the paper is accepted, and the training code and network weights will be published immediately.

We've desinged a evolutionary neural architecture search algorithm for generative adversarial networks (GANs), dubbed T-EAGAN. Experiments validate the effectiveness of T-EAGAN on the task of unconditional image generation. Extensive experiments on the CIFAR-10 and STL-10 datasets demonstrated that T-EAGAN only requires 1.08 GPU days to find out a superior GAN architecture in a search space including approximately 10¹⁵ network architectures. Our best found network architectures could outperform those obtained by other neural architecture search methods with the performance metric results(IS=8.957±0.08, FID=9.432) on CIFAR-10 and (IS=10.576±0.085, FID=20.323) on STL-10.

Fig:framework for T-EAGAN

picture1

picture2

The search environment is consistent with AlphaGAN，to run this code, you need:

• PyTorch 1.3.0
• TensorFlow 1.15.0
• cuda 10.0

Other requirements are in environment.yaml

conda env create -f environment.yaml

you need to create "fid_stat" directory and download the statistical files of real images.

mkdir fid_stat

bash EAGAN_Only_G30.sh

bash ./scripts/train_arch_cifar10.sh

bash ./scripts/train_arch_stl10.sh

bash ./scripts/test_arch_cifar10.sh

bash ./scripts/test_arch_stl10.sh

Some of the codes are built by:

1.EAGAN

2.AlphaGAN

3.Inception Score code from OpenAI's Improved GAN (official).

4.FID Score code and CIFAR-10 statistics file from (official).

Thanks them for their great works!