Research map / Architectures and efficiency / CNN design
Sparsity Aware Normalization 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 |
|---|---|---|---|
| Replace spectral normalization with sparsity-aware normalization | Standard CNN GAN · CIFAR-10 | Inception score: 7.58 ± .12 -> 7.89 ± .09 | Helped |
| Replace spectral normalization with sparsity-aware normalization | Standard CNN GAN · STL-10 | Inception score: 8.79 ± .14 -> 9.18 ± .06 | Helped |
| Replace spectral normalization with sparsity-aware normalization | ResNet GAN · CIFAR-10 | Inception score: 8.22 ± .05 -> 8.43 ± .13 | Helped |
| Replace spectral normalization with sparsity-aware normalization | ResNet GAN · STL-10 | Inception score: 9.10 ± .04 -> 9.21 ± .10 | Helped |
| Replace BigGAN critic spectral normalization with SAN | BigGAN · CIFAR-10 conditional generation | Inception score: 9.24 ± .16 -> 9.53 ± .13 | Helped |
| Apply SAN to SPADE discriminator | SPADE · Cityscapes label-to-image translation | FID: 63.65 -> 58.56 | Helped |
| Use SAN with a compact SRGAN generator | SRGAN generator · BSD100 4× super-resolution | PSNR: 25.69 -> 26.15 | Helped |
| Use SAN with a compact SRGAN generator | SRGAN generator · BSD100 4× super-resolution | NIQE: 3.56 -> 3.21 | Helped |
| Use SAN instead of the StarGAN baseline normalization | StarGAN · CelebA gender translation | FID: 85.10 -> 71.41 | Helped |
| Normalize only once every 1000 critic steps (ablation) | SAN-GAN · CIFAR-10 | Inception score | Helped |
| Set the normalization correction to 1.1 instead of the proposed 1.0 (ablation) | SAN-GAN · CIFAR-10 | Inception score | Helped |
| Use SAN rather than spectral normalization across varied training hyperparameters (ablation) | CIFAR-10 GAN · CIFAR-10, hyperparameter settings A-F | Inception score | Mixed |
Closest papers by meaning
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.93)
- Spectral Normalization for Generative Adversarial Networks (ICLR 2018 · similarity 0.92)
- Why Spectral Normalization Stabilizes GANs: Analysis and Improvements (NeurIPS 2021 · similarity 0.92)
- CHAIN: Enhancing Generalization in Data-Efficient GANs via lipsCHitz continuity constrAIned Normalization (CVPR 2024 · similarity 0.89)
- SAN: Inducing Metrizability of GAN with Discriminative Normalized Linear Layer (ICLR 2024 · similarity 0.88)
- RG-GAN: Dynamic Regenerative Pruning for Data-Efficient Generative Adversarial Networks (AAAI 2024 · similarity 0.88)
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.