Research map / Image generation / GANs
Re-GAN: Data-Efficient GANs Training via Architectural Reconfiguration
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 |
|---|---|---|---|
| Reconfigure generator connections during training | StyleGAN2 · CIFAR-10, 10% data | FID: 36.45 -> 32.91 | Helped |
| Reconfigure generator connections during training | StyleGAN2 · Tiny-ImageNet, 10% data | FID: 84.86 -> 82.63 | Helped |
| Reconfigure generator connections during training | ProGAN · CIFAR-10, full data | FID: 26.14 -> 21.17 | Helped |
| Reconfigure generator connections during training | AutoGAN (Top B) · CIFAR-10, full data | FID: 18.08 -> 17.30 | Helped |
| Use the same learning rate in sparse and dense phases (ablation) | Re-SNGAN · CIFAR-10, full data | FID: 17.87 -> 19.06 | Hurt |
| Randomly initialize weights when growing connections (ablation) | Re-SNGAN · CIFAR-10, full data | FID: 17.87 -> 19.71 | Hurt |
| Use 1/10 the initial learning rate during network growing (ablation) | Re-SNGAN · CIFAR-10, full data | FID: 19.06 -> 17.87 | Helped |
| Apply pruning only to the generator rather than to both networks (ablation) | SNGAN · CIFAR-10, full data | FID | Helped |
| Prune both generator and discriminator during training (ablation) | SNGAN · CIFAR-10, full data | FID | No clear effect |
| Train with Re-SNGAN rather than SNGAN | SNGAN · CIFAR-10, 10%–100% training data | FID | Helped |
| Use Re-GAN with DiffAug | StyleGAN2 · Few-shot datasets, 256×256 | FID | Helped |
| Reconfigure generator connections during training | StyleGAN2 · FFHQ, 70k images, 1024×1024 | FID: 4.35 -> 4.19 | Helped |
Closest papers by meaning
- RG-GAN: Dynamic Regenerative Pruning for Data-Efficient Generative Adversarial Networks (AAAI 2024 · similarity 0.92)
- The GAN is dead; long live the GAN! A Modern GAN Baseline (NeurIPS 2024 · similarity 0.92)
- A Large-Scale Study on Regularization and Normalization in GANs (ICML 2019 · similarity 0.92)
- Projected GANs Converge Faster (NeurIPS 2021 · similarity 0.91)
- Data-Efficient GAN Training Beyond (Just) Augmentations: A Lottery Ticket Perspective (NeurIPS 2021 · similarity 0.91)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 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.