Research map / Image generation / GANs
Improved Consistency Regularization for GANs
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 balanced consistency regularization by augmenting generated images as well as real images | GAN with vanilla consistency regularization · CIFAR-10 conditional | FID: 11.48 -> 10.54 | Helped |
| Add latent consistency regularization, including latent perturbations and generator sensitivity encouragement | BigGAN · ImageNet-2012 conditional | FID: 6.66 -> 5.87 | Helped |
| Combine balanced and latent consistency regularization | BigGAN · CIFAR-10 conditional | FID: 11.48 -> 9.21 | Helped |
| Combine balanced and latent consistency regularization | BigGAN · ImageNet-2012 conditional | FID: 6.66 -> 5.38 | Helped |
| Combine balanced and latent consistency regularization | DCGAN · CIFAR-10 unconditional | FID: 18.72 -> 15.87 | Helped |
| Combine balanced and latent consistency regularization | ResNet GAN · CIFAR-10 unconditional | FID: 14.56 -> 13.36 | Helped |
| Combine balanced and latent consistency regularization | DCGAN · CelebA unconditional | FID: 16.97 -> 15.43 | Helped |
| Add generated-image consistency regularization to vanilla CR (ablation) | DCGAN · CIFAR-10 with 32x32 cutout | FID: 38.03 ± 1.01 -> 17.94 ± 0.48 | Helped |
| Increase lambda_fake for balanced consistency regularization (ablation) | DCGAN · CIFAR-10 with 32x32 cutout | artifact fraction: 0.66 ± 0.03 -> 0.01 ± 0.01 | Helped |
| Set lambda_fake to zero, removing consistency regularization on generated images (ablation) | balanced consistency regularization · CIFAR-10 DCGAN, hinge loss | FID | Hurt |
| Use a moderate generator coefficient, lambda_gen = 0.5 (ablation) | latent consistency regularization · CIFAR-10 ResNet GAN | FID | Helped |
| Use latent perturbations with standard deviation sigma_noise = 0.07 (ablation) | latent consistency regularization · CIFAR-10 ResNet GAN | FID | Helped |
Closest papers by meaning
- Consistency Regularization for Generative Adversarial Networks (ICLR 2020 · similarity 0.96)
- NICE: NoIse-modulated Consistency rEgularization for Data-Efficient GANs (NeurIPS 2023 · similarity 0.94)
- Consistency-GAN: Training GANs with Consistency Model (AAAI 2024 · similarity 0.91)
- Stabilizing Training of Generative Adversarial Networks through Regularization (NeurIPS 2017 · similarity 0.91)
- SMaRt: Improving GANs with Score Matching Regularity (ICML 2024 · similarity 0.89)
- Style Quantization for Data-Efficient GAN Training (CVPR 2025 · similarity 0.89)
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.