Research map / Learning theory and algorithms / Optimal transport
Towards Generalized Implementation of Wasserstein Distance 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 |
|---|---|---|---|
| Relax the critic constraint to Sobolev duality, using augmented Lagrangian regularization instead of the WGAN-AL Lipschitz constraint | WGAN-AL · CIFAR-10 | FID: 17.86 -> 16.75 | Helped |
| Relax the critic constraint to Sobolev duality, using augmented Lagrangian regularization instead of the WGAN-AL Lipschitz constraint | WGAN-AL · Tiny-ImageNet | FID: 18.70 -> 18.32 | Helped |
| Relax the critic constraint to Sobolev duality, using gradient penalty instead of the WGAN-GP Lipschitz constraint | WGAN-GP · CIFAR-10 | IS: 7.85 -> 7.98 | Helped |
| Relax the critic constraint to Sobolev duality, using gradient penalty instead of the WGAN-GP Lipschitz constraint | WGAN-GP · Tiny-ImageNet | FID: 18.70 -> 18.50 | Helped |
| Increase interpolation sampling size to m=8 (ablation) | WGAN-GP · CIFAR-10 | IS: 7.85 -> 7.88 | Helped |
| Increase interpolation sampling size to m=8 (ablation) | WGAN-AL · CIFAR-10 | FID: 17.86 -> 17.52 | Helped |
| Increase interpolation sampling size to m=8 (ablation) | SGAN with μ=μ_GP · Tiny-ImageNet | FID: 18.90 -> 18.75 | Helped |
| Use augmented Lagrangian instead of gradient penalty for SWGAN (ablation) | SWGAN · CIFAR-10 | FID: 17.50 -> 16.75 | Helped |
| Use augmented Lagrangian instead of gradient penalty for SWGAN (ablation) | SWGAN · CIFAR-10 | IS: 7.98 -> 7.93 | Hurt |
| Use augmented Lagrangian instead of gradient penalty for SWGAN (ablation) | SWGAN · Tiny-ImageNet | IS: 8.38 -> 8.41 | Helped |
| Use augmented Lagrangian instead of gradient penalty for SWGAN (ablation) | SWGAN · Tiny-ImageNet | FID: 18.50 -> 18.32 | Helped |
| Replace the standard SWGAN gradient penalty with zero-centered penalty (ablation) | SWGAN-GP · Real-world image generation | sample quality | No clear effect |
Closest papers by meaning
- A Two-Step Computation of the Exact GAN Wasserstein Distance (ICML 2018 · similarity 0.93)
- Lipschitz Generative Adversarial Nets (ICML 2019 · similarity 0.92)
- Improving the Improved Training of Wasserstein GANs: A Consistency Term and Its Dual Effect (ICLR 2018 · similarity 0.92)
- Banach Wasserstein GAN (NeurIPS 2018 · similarity 0.91)
- Hidden Convexity of Wasserstein GANs: Interpretable Generative Models with Closed-Form Solutions (ICLR 2022 · similarity 0.90)
- Improved Training of Wasserstein GANs (NeurIPS 2017 · 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.