Research map / Image generation / GANs
On Stabilizing Generative Adversarial Training With Noise
Check 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 distribution filtering to SNGAN | SNGAN · CIFAR-10 | FID: 29.1±0.4 -> 25.9±0.3 | Helped |
| Add distribution filtering to SNGAN | SNGAN · CIFAR-10 | IS: 7.26±.06 -> 7.47±.08 | Helped |
| Add distribution filtering to SNGAN | SNGAN · CelebA | FID: 13.2±0.3 -> 10.5±0.4 | Helped |
| Add fixed Gaussian distribution filtering | Progressive GAN · CIFAR-10 | FID: 29.4 -> 26.8 | Helped |
| Train on both clean and noisy examples rather than noisy examples only (ablation) | GAN · CIFAR-10 | FID: 94.9±4.9 -> 29.7±0.6 | Helped |
| Train on both clean and noisy examples rather than noisy examples only (ablation) | GAN · STL-10 | FID: 107.9±2.3 -> 66.5±2.3 | Helped |
| Add clean examples alongside learned-noise examples (ablation) | GAN · CIFAR-10 | FID: 69.0±3.4 -> 28.8±0.7 | Helped |
| Anneal Gaussian noise toward zero while feeding only noisy examples (ablation) | GAN · CIFAR-10 | FID: 46.1±0.7 -> 44.5±3.2 | Helped |
| Feed only fixed Gaussian-noisy examples instead of clean and noisy examples (ablation) | GAN · CIFAR-10 | FID: 29.7±0.6 -> 94.9±4.9 | Hurt |
| Feed only learned-noise examples instead of clean and noisy examples (ablation) | GAN · CIFAR-10 | FID: 28.8±0.7 -> 69.0±3.4 | Hurt |
| Increase the noise penalty weight from λ=1 to λ=10 (ablation) | DFGAN · CIFAR-10 | FID: 26.5±0.6 -> 29.8±0.4 | Hurt |
| Use the faster alternative mini-batch construction (ablation) | DFGAN (λ=1) · CIFAR-10 | FID: 26.5±0.6 -> 28.7±0.6 | Hurt |
Closest papers by meaning
- Unrolled Generative Adversarial Networks (ICLR 2017 · similarity 0.93)
- Stabilizing Training of Generative Adversarial Networks through Regularization (NeurIPS 2017 · similarity 0.92)
- Training Generative Adversarial Networks with Limited Data (NeurIPS 2020 · similarity 0.92)
- Improving Generalization and Stability of Generative Adversarial Networks (ICLR 2019 · similarity 0.92)
- Smoothness and Stability in GANs (ICLR 2020 · similarity 0.91)
- Consistency-GAN: Training GANs with Consistency Model (AAAI 2024 · similarity 0.90)
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.