@inproceedings{bb82bc66b6124c4fa03c7c78c7cc635f,
title = "EvolGAN: Evolutionary Generative Adversarial Networks",
abstract = "We propose to use a quality estimator and evolutionary methods to search the latent space of generative adversarial networks trained on small, difficult datasets, or both. The new method leads to the generation of significantly higher quality images while preserving the original generator{\textquoteright}s diversity. Human raters preferred an image from the new version with frequency 83.7% for Cats, 74% for FashionGen, 70.4% for Horses, and 69.2% for Artworks - minor improvements for the already excellent GANs for faces. This approach applies to any quality scorer and GAN generator.",
author = "Baptiste Roziere and Fabien Teytaud and Vlad Hosu and Hanhe Lin and Jeremy Rapin and Mariia Zameshina and Olivier Teytaud",
note = "Funding Information: Acknowledgments. Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation), Project-ID 251654672, TRR 161 (Project A05). Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 15th Asian Conference on Computer Vision, ACCV 2020 ; Conference date: 30-11-2020 Through 04-12-2020",
year = "2021",
doi = "10.1007/978-3-030-69538-5_41",
language = "English",
isbn = "978-3-030-69537-8",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Nature Switzerland AG",
pages = "679--694",
editor = "Hiroshi Ishikawa and Cheng-Lin Liu and Tomas Pajdla and Jianbo Shi",
booktitle = "Computer Vision – ACCV 2020",
address = "Switzerland",
}