GLeaD: Improving GANs with A Generator-Leading Task
Qingyan Bai1     Ceyuan Yang2     Yinghao Xu3     Xihui Liu4     Yujiu Yang1     Yujun Shen5
1 Tsinghua University     2 Shanghai AI Laboratory     3 The Chinese University of Hong Kong    
4 The University of Hong Kong      5 Ant Group
Overview
This work aims at improving Generative adversarial network (GAN) with a generator-leading task. GAN is formulated as a two-player game between a generator (G) and a discriminator (D), where D is asked to differentiate whether an image comes from real data or is produced by G. Under such a formulation, D plays as the rule maker and hence tends to dominate the competition. Towards a fairer game in GANs, we propose a new paradigm for adversarial training, which makes G assign a task to D as well. Specifically, given an image, we expect D to extract representative features that can be adequately decoded by G to reconstruct the input. That way, instead of learning freely, D is urged to align with the view of G for domain classification.
Results
Quantitative comparisons on FFHQ, LSUN Bedroom and LSUN Church. P and R in the table respectively denote precision and recall. Our method improves StyleGAN2 in large datasets in terms of FID and recall. Combined with GGDR (Ours* in the table), GLeaD could further introduce significant gains, achieving new state-of-the-art performance on various datasets.
Synthesized images by our models respectively trained on FFHQ, LSUN Bedroom and Church.
Reconstruction results of real and synthesized input images. These results indicate that our D could learn features aligned with the domain of G, matching our motivation.
BibTeX
@article{bai2022glead,
  title   = {GLeaD: Improving GANs with A Generator-Leading Task},
  author  = {Bai, Qingyan and Yang, Ceyuan and Xu, Yinghao and Liu, Xihui and Yang, Yujiu and Shen, Yujun},
  journal = {arXiv preprint arXiv:2212.03752},
  year    = {2022}
}
Related Work
Jeff Donahue, Philipp Krähenbühl, Trevor Darrell. Adversarial Feature Learning. ICLR 2017.
Comment: Proposes to learn an additional encoder to project real samples back into GAN's latent space, and discriminate samples jointly in data and latent space.