Interpreting the Latent Space of GANs for Semantic Face Editing
Yujun Shen1  Jinjin Gu2  Xiaoou Tang1  Bolei Zhou1 
1The Chinese University of Hong Kong
2The Chinese University of Hong Kong, Shenzhen
Overview
We find that the latent code for well-trained generative models, such as PGGAN and StyleGAN, actually learns a disentangled representation after some linear transformations. Based on our analysis, we propose a simple and general technique, called InterFaceGAN, for semantic face editing in latent space. We manage to control the pose as well as other facial attributes, such as gender, age, eyeglasses. More importantly, we are able to correct the artifacts made by GANs.
Results
We manipulate the following attributes with PGGAN.
Pose Age Gender
Expression Eyeglasses Artifacts
Check more results in the following video.
Bibtex
@inproceedings{shen2019interpreting,
  title     = {Interpreting the Latent Space of GANs for Semantic Face Editing},
  author    = {Shen, Yujun and Gu, Jinjin and Tang, Xiaoou and Zhou, Bolei},
  booktitle = {CVPR},
  year      = {2020}
}
Related Work
D. Bau, JY. Zhu, H. Strobelt, B. Zhou, JB. Tenenbaum, WT. Freeman, A. Torralba. GAN Dissection: Visualizing and Understanding Generative Adversarial Networks. ICLR 2019.
Comment: Dissects neurons in GANs from the perspective of spatial feature map.
L. Goetschalckx, A. Andonian, A. Oliva, P. Isola. GANalyze: Toward Visual Definitions of Cognitive Image Properties. ICCV, 2019.
Comment: Navigates the manifold of GAN's latent space to increase memorability.
A. Jahanian, L. Chai, P. Isola. On the "Steerability" of Generative Adversarial Networks. ICLR, 2020.
Comment: Shifts the data distribution by steering the latent code to fit camera movements and color changes.
C. Yang, Y. Shen, B. Zhou. Semantic Hierarchy Emerges in Deep Generative Representations for Scene Synthesis. arXiv preprint arXiv:1911.09267.
Comment: Explores the emergent semantic hierarchy in scene synthesis models.