[1912.04958] Analyzing and Improving the Image Quality of StyleGANcontact arXivarXiv Twitter

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 generator normalization, revisit progressive growing, and regularize the generator to encourage good conditioning in the mapping from latent vectors 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 detect if an image is generated by 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

2 mentions: @shion_honda
Keywords: stylegan
Date: 2020/02/12 20:21

Referring Tweets

@shion_honda StyleGAN2 [Karras+, 2019] StyleGANの水滴状のノイズや特徴の位置特異性を指摘。生成器を見直し、PPLが画像の質を反映することから正則化を見直し、progressive growingを取り除き全体を同時に訓練することで問題を解決。PPL正則化により逆変換が簡単になった。 t.co/jiTwP4hSSo #NowReading t.co/5wBq1IpXtZ

Related Entries

Read more StyleGAN: A Style-Based Generator Architecture for Generative Adversarial Networksの実装 - 機械学習奮闘記
5 users, 2 mentions 2019/03/06 11:14
Read more StyleGAN解説 CVPR2019読み会@DeNA
0 users, 0 mentions 2019/08/26 08:17
Read more Making Anime Faces With StyleGAN - Gwern.net
2 users, 14 mentions 2019/03/21 02:17
Read more GitHub - Puzer/stylegan-encoder: StyleGAN Encoder - converts real images to latent space
0 users, 4 mentions 2019/02/20 18:48