Research map / Architectures and efficiency / RNNs and normalization
Why Spectral Normalization Stabilizes GANs: Analysis and Improvements
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 scaling to SN weights (ablation) | SN · CIFAR10 | IS: 7.12 $\pm$ 0.07 -> 7.38 $\pm$ 0.06 | Helped |
| add bidirectional normalization to SN (ablation) | SN · CIFAR10 | FID: 31.43 $\pm$ 0.90 -> 26.94 $\pm$ 0.58 | Helped |
| add bidirectional normalization to SN (ablation) | SN · STL10 | FID: 44.35 $\pm$ 0.54 -> 42.98 $\pm$ 0.54 | Helped |
| add bidirectional normalization to SN (ablation) | SN · CelebA | FID: 9.43 $\pm$ 0.09 -> 9.05 $\pm$ 0.13 | Helped |
| add bidirectional normalization to SN (ablation) | SN · ILSVRC2012 | IS: 12.84 $\pm$ 0.33 -> 1.77 $\pm$ 0.13 | Hurt |
| add scaling to BSN (ablation) | BSN · ILSVRC2012 | IS: 1.77 $\pm$ 0.13 -> 13.23 $\pm$ 0.16 | Helped |
| add scaling to BSN (ablation) | BSN · CIFAR10 | IS: 7.54 $\pm$ 0.04 -> 7.54 $\pm$ 0.04 | No clear effect |
| add bidirectional normalization and scaling (ablation) | SN · ILSVRC2012 | FID: 75.06 $\pm$ 2.38 -> 69.04 $\pm$ 1.46 | Helped |
Closest papers by meaning
- Spectral Normalization for Generative Adversarial Networks (ICLR 2018 · similarity 0.94)
- Sparsity Aware Normalization for GANs (AAAI 2021 · similarity 0.92)
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.90)
- On Computation and Generalization of Generative Adversarial Networks under Spectrum Control (ICLR 2019 · similarity 0.90)
- Generalizable Adversarial Training via Spectral Normalization (ICLR 2019 · similarity 0.89)
- CHAIN: Enhancing Generalization in Data-Efficient GANs via lipsCHitz continuity constrAIned Normalization (CVPR 2024 · 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.