Research map / Image generation / GANs
Improving MMD-GAN Training with Repulsive Loss Function
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 the attractive discriminator loss with the repulsive loss | MMD-GAN · CIFAR-10 | FID: 28.38 -> 16.65 | Helped |
| Replace the attractive discriminator loss with the repulsive loss | MMD-GAN · STL-10 | FID: 57.52 -> 36.67 | Helped |
| Replace the attractive discriminator loss with the repulsive loss | MMD-GAN · CelebA | FID: 13.03 -> 7.20 | Helped |
| Replace the attractive discriminator loss with the repulsive loss | MMD-GAN · LSUN bedroom | FID | Helped |
| Use a bounded RBF kernel instead of the five-kernel attractive MMD-rbf setup (ablation) | MMD-GAN · CIFAR-10 | FID: 28.38 -> 25.25 | Helped |
| Use a bounded RBF kernel instead of the five-kernel attractive MMD-rbf setup (ablation) | MMD-GAN · STL-10 | FID: 57.52 -> 51.86 | Helped |
| Use a bounded RBF kernel instead of the five-kernel attractive MMD-rbf setup (ablation) | MMD-GAN · CelebA | FID: 13.03 -> 10.09 | Helped |
| Add the bounded RBF kernel to the repulsive loss (ablation) | MMD-rep · CIFAR-10 | FID: 16.65 -> 16.21 | Helped |
| Add the bounded RBF kernel to the repulsive loss (ablation) | MMD-rep · STL-10 | FID: 36.67 -> 37.63 | Hurt |
| Add the bounded RBF kernel to the repulsive loss (ablation) | MMD-rep · CelebA | FID: 7.20 -> 6.79 | Helped |
| Add the bounded RBF kernel to the repulsive loss (ablation) | MMD-rep · LSUN bedroom | FID: 16.91 -> 12.52 | Helped |
| Increase the discriminator-loss parameter from -1 to -0.5 (ablation) | MMD-GAN with RBF kernel · CIFAR-10 | FID | Helped |
Closest papers by meaning
- Demystifying MMD GANs (ICLR 2018 · similarity 0.91)
- On gradient regularizers for MMD GANs (NeurIPS 2018 · similarity 0.89)
- Adaptive Weighted Discriminator for Training Generative Adversarial Networks (CVPR 2021 · similarity 0.88)
- Generative Models and Model Criticism via Optimized Maximum Mean Discrepancy (ICLR 2017 · similarity 0.88)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.88)
- Improving the Training of Data-Efficient GANs via Quality Aware Dynamic Discriminator Rejection Sampling (CVPR 2025 · similarity 0.87)
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.