Research map / Image generation / GANs
Consistency Regularization for Generative Adversarial Networks
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 |
|---|---|---|---|
| Add consistency regularization to SNDCGAN | SNDCGAN · CIFAR-10, unconditional | FID: 24.73 -> 18.72 | Helped |
| Add consistency regularization to ResNet | ResNet · CIFAR-10, unconditional | FID: 19.00 -> 14.56 | Helped |
| Add consistency regularization to SNDCGAN | SNDCGAN · CelebA, unconditional | FID: 25.95 -> 16.97 | Helped |
| Add consistency regularization to BigGAN* | BigGAN* · CIFAR-10, class-conditional | FID: 20.42 to 11.48 | Helped |
| Add consistency regularization to BigGAN* | BigGAN* · ImageNet-2012, class-conditional | FID: 7.75 to 6.66 | Helped |
| Use Gaussian noise augmentation (ablation) | CR-GAN · CIFAR-10 | FID: 21.91$\pm0.32$ | Hurt |
| Use random shift and flip augmentation (ablation) | CR-GAN · CIFAR-10 | FID: 16.04$\pm0.17$ | Helped |
| Use cutout augmentation (ablation) | CR-GAN · CIFAR-10 | FID: 17.10$\pm0.29$ | Mixed |
| Add cutout to random shift and flip (ablation) | CR-GAN · CIFAR-10 | FID: 19.46$\pm0.26$ | Hurt |
| Add consistency regularization on the final discriminator layer rather than the first few intermediate layers (ablation) | CR-GAN · CIFAR-10 | FID | Helped |
| Add consistency regularization to generated samples as well as real samples (ablation) | CR-GAN · CIFAR-10, SNDCGAN | FID: 20.21 of CR-Real to 15.51 | Helped |
| Add consistency regularization to generated samples as well as real samples (ablation) | CR-GAN · CIFAR-10, ResNet | FID: 14.93 vs 15.07 | Hurt |
Closest papers by meaning
- Improved Consistency Regularization for GANs (AAAI 2021 · similarity 0.96)
- Stabilizing Training of Generative Adversarial Networks through Regularization (NeurIPS 2017 · similarity 0.94)
- NICE: NoIse-modulated Consistency rEgularization for Data-Efficient GANs (NeurIPS 2023 · similarity 0.93)
- Which Training Methods for GANs do actually Converge? (ICML 2018 · similarity 0.91)
- Consistency-GAN: Training GANs with Consistency Model (AAAI 2024 · similarity 0.91)
- Mode Regularized Generative Adversarial Networks (ICLR 2017 · 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.