Research map / Label-efficient and robust learning / Recognition and long-tail
A Large-Scale Study on Regularization and Normalization in 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 spectral normalization | SNDCGAN · CIFAR10 | FID: 29.75 -> 26.66 | Helped |
| Add spectral normalization | SNDCGAN · CelebA-HQ-128 | FID: 66.42 -> 31.14 | Helped |
| Add spectral normalization | SNDCGAN · LSUN-bedroom | FID: 180.36 -> 63.46 | Helped |
| Add spectral normalization | ResNet19 · LSUN-bedroom | FID: 160.97 -> 70.30 | Helped |
| Add layer normalization | ResNet19 · CelebA-HQ-128 | FID: 43.73 -> 32.46 | Helped |
| Add spectral normalization | ResNet CIFAR · CIFAR10 | FID: 31.40 -> 23.57 | Helped |
| Add gradient penalty to spectral normalization | SNDCGAN · CIFAR10 | FID: 26.66 -> 26.23 | Helped |
| Add gradient penalty to spectral normalization | ResNet CIFAR · CIFAR10 | FID: 23.57 -> 22.98 | Helped |
| Add batch normalization to the discriminator | ResNet19 GAN · CelebA-HQ-128 and LSUN-bedroom | FID | Hurt |
| Use gradient penalty rather than no regularizer | ResNet19 GAN · CelebA-HQ-128 and LSUN-bedroom | FID | Mixed |
| Use spectral normalization rather than no normalization | GANs with ResNet19 or SNDCGAN · CelebA-HQ-128 and LSUN-bedroom | FID | Helped |
| Combine gradient penalty with spectral or layer normalization | ResNet19 GAN · CelebA-HQ-128 and LSUN-bedroom | FID | Helped |
Closest papers by meaning
- Are GANs Created Equal? A Large-Scale Study (NeurIPS 2018 · similarity 0.93)
- Stabilizing Training of Generative Adversarial Networks through Regularization (NeurIPS 2017 · similarity 0.93)
- Sparsity Aware Normalization for GANs (AAAI 2021 · similarity 0.93)
- Variational Annealing of GANs: A Langevin Perspective (ICML 2019 · similarity 0.93)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · similarity 0.91)
- MS$^3$D: A RG Flow-Based Regularization for GAN Training with Limited Data (ICML 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.