Analyzing and Improving the Image Quality of StyleGAN

被引:3812
作者
Karras, Tero [1 ]
Laine, Samuli [1 ]
Aittala, Miika [1 ]
Hellsten, Janne [1 ]
Lehtinen, Jaakko [1 ,2 ]
Aila, Timo [1 ]
机构
[1] NVIDIA, Santa Clara, CA 95051 USA
[2] Aalto Univ, Espoo, Finland
来源
2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2020) | 2020年
关键词
D O I
10.1109/CVPR42600.2020.00813
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
The style-based GAN architecture (StyleGAN) yields state-of-the-art results in data-driven unconditional generative image modeling. We expose and analyze several of its characteristic artifacts, and propose changes in both model architecture and training methods to address them. In particular, we redesign the generator normalization, revisit progressive growing, and regularize the generator to encourage good conditioning in the mapping from latent codes to images. In addition to improving image quality, this path length regularizer yields the additional benefit that the generator becomes significantly easier to invert. This makes it possible to reliably attribute a generated image to a particular network. We furthermore visualize how well the generator utilizes its output resolution, and identify a capacity problem, motivating us to train larger models for additional quality improvements. Overall, our improved model redefines the state of the art in unconditional image modeling, both in terms of existing distribution quality metrics as well as perceived image quality.
引用
收藏
页码:8107 / 8116
页数:10
相关论文
共 45 条
[1]   Image2StyleGAN: How to Embed Images Into the StyleGAN Latent Space? [J].
Abdal, Rameen ;
Qin, Yipeng ;
Wonka, Peter .
2019 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2019), 2019, :4431-4440
[2]  
Albright M., 2019, CVPR WORKSH
[3]  
Bergstrom Carl, Which face is real?
[4]  
Brock A, 2019, Arxiv, DOI arXiv:1809.11096
[5]  
Denton E, 2015, Arxiv, DOI arXiv:1506.05751
[6]  
Dumoulin V, 2017, Arxiv, DOI [arXiv:1610.07629, DOI 10.48550/ARXIV.1610.07629]
[7]  
Gabbay A, 2019, Arxiv, DOI [arXiv:1906.11880, DOI 10.48550/ARXIV.1906.11880]
[8]  
Geirhos R, 2019, Arxiv, DOI arXiv:1811.12231
[9]  
Ghiasi G, 2017, Arxiv, DOI [arXiv:1705.06830, DOI 10.48550/ARXIV.1705.06830]
[10]  
Glorot X., 2010, P 13 INT C ART INT S, P249, DOI DOI 10.1109/LGRS.2016.2565705