Research map / Image generation / GANs
Generative Modeling Using the Sliced Wasserstein Distance
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 |
|---|---|---|---|
| Train the generator by minimizing sliced Wasserstein distance instead of using the GAN or WGAN objective | Generative modeling on MNIST · MNIST | training stability | Helped |
| Train the generator with sliced Wasserstein distance rather than the GAN objective | Generative modeling on MNIST · MNIST | sample diversity | Helped |
| Add a discriminator feature representation and match projected distributions in that representation | SWG for high-dimensional data · CIFAR-10, LSUN Bedrooms, CelebA | training success across architectures | Helped |
| Use 5000 rather than 10000 random projections (ablation) | SWG generator updates · Generator update, batch size 64 | time (s): 0.072 -> 0.06 | Helped |
| Use 5000 rather than 10000 random projections (ablation) | SWG generator updates · Generator update, batch size 256 | time (s): 0.17 -> 0.146 | Helped |
| Use convolutional rather than fully connected generator layers (ablation) | SWG on MNIST · MNIST | training success | No clear effect |
| Add batch normalization to the generator (ablation) | SWG on MNIST · MNIST | training success | No clear effect |
| Use a convolutional generator rather than a fully connected generator (ablation) | GAN and WGAN baselines · MNIST | training stability | Mixed |
| Add batch normalization to the generator (ablation) | GAN and WGAN baselines · MNIST | training stability | Mixed |
| Use a DCGAN discriminator with 64 filters in each layer (ablation) | SWG · CIFAR-10, LSUN Bedrooms, CelebA | training success | No clear effect |
| Double the number of filters in each discriminator layer (ablation) | SWG · CIFAR-10, LSUN Bedrooms, CelebA | training success | No clear effect |
| Add two stride-1 convolutional layers to the generator (ablation) | SWG · CIFAR-10, LSUN Bedrooms, CelebA | training success | No clear effect |
Closest papers by meaning
- Max-Sliced Wasserstein Distance and Its Use for GANs (CVPR 2019 · similarity 0.94)
- A Two-Step Computation of the Exact GAN Wasserstein Distance (ICML 2018 · similarity 0.93)
- Don't let your Discriminator be fooled (ICLR 2019 · similarity 0.92)
- Sliced Wasserstein Generative Models (CVPR 2019 · similarity 0.92)
- Banach Wasserstein GAN (NeurIPS 2018 · similarity 0.92)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.91)
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.