Research map / Image generation / GANs
Progressive Growing of GANs for Improved Quality, Stability, and Variation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| progressively grow the generator and discriminator from low to high resolution | GAN training · CelebA and LSUN bedroom 128² | SWD | Helped |
| progressively grow the generator and discriminator from low to high resolution | GAN training · CelebA up to 1024² | training speed: roughly a $5.4\times$ speedup | Helped |
| reduce minibatch size from 64 to 16 (ablation) | progressive GAN training · CelebA 128² | SWD | Hurt |
| adjust hyperparameters and remove batch and layer normalization (ablation) | progressive GAN training with minibatch size 16 · CelebA 128² | image quality | Helped |
| add minibatch discrimination (ablation) | progressive GAN · CelebA 128² | SWD | No clear effect |
| add minibatch standard deviation (ablation) | progressive GAN · CelebA 128² | SWD | Helped |
| add equalized learning rate and pixelwise feature normalization (ablation) | GAN training · MNIST-1K discrete mode test | KL divergence | Helped |
| add minibatch standard deviation (ablation) | GAN training with equalized learning rate and pixelwise normalization · MNIST-1K discrete mode test | KL divergence | Helped |
| progressively grow the networks (ablation) | GAN training · MNIST-1K discrete mode test | KL divergence | No clear effect |
| use WGAN-GP instead of the original GAN loss (ablation) | GAN training with the networks specified by Metz et al. · MNIST-1K discrete mode test | mode coverage | Helped |
| use LSGAN instead of WGAN-GP (ablation) | Progressive GAN · CelebA 1024² | training stability | Hurt |
| increase the WGAN-GP regularization parameter to $\gamma=750$ (ablation) | Progressive GAN · CIFAR10 | inception score | Helped |
Closest papers by meaning
- Projected GANs Converge Faster (NeurIPS 2021 · similarity 0.92)
- Dynamically Grown Generative Adversarial Networks (AAAI 2021 · similarity 0.92)
- Large Scale GAN Training for High Fidelity Natural Image Synthesis (ICLR 2019 · similarity 0.92)
- Training Generative Adversarial Networks with Limited Data (NeurIPS 2020 · similarity 0.91)
- Improved Techniques for Training GANs (NeurIPS 2016 · similarity 0.91)
- Progressive Augmentation of GANs (NeurIPS 2019 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.