Research map / Image generation / GANs
On GANs and GMMs
Eitan Richardson, Yair Weiss
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Use maximum-likelihood MFA rather than WGAN-GP | CelebA generative modeling · CelebA | NDB/K: 0.42 -> 0.21 (K=100); 0.32 -> 0.12 (K=200); 0.27 -> 0.16 (K=300) | Improved |
| Use maximum-likelihood MFA rather than WGAN | MNIST generative modeling · MNIST | NDB/K: 0.16 -> 0.14 (K=100); 0.20 -> 0.13 (K=200); 0.21 -> 0.14 (K=300) | Improved |
| Use maximum-likelihood MFA rather than WGAN | SVHN generative modeling · SVHN | NDB/K: 0.87 -> 0.32 (K=100); 0.83 -> 0.23 (K=200); 0.82 -> 0.24 (K=300) | Improved |
| Add pix2pix sharpening to MFA samples (ablation) | MFA · CelebA | NDB/K: 0.21 -> 0.34 (K=100); 0.12 -> 0.34 (K=200); 0.16 -> 0.33 (K=300) | Hurt |
| Add pix2pix sharpening to MFA samples (ablation) | MFA · CelebA | sharpness: -5.4 -> -3.5 | Improved |
| Train MFA adversarially instead of by maximum likelihood (ablation) | MFA · CelebA | NDB/K: 0.21 -> 0.33 (K=100); 0.12 -> 0.30 (K=200); 0.16 -> 0.22 (K=300) | Hurt |
| Train MFA adversarially instead of by maximum likelihood (ablation) | MFA · CelebA | sharpness: -5.4 -> -3.8 | Improved |
| Increase the number of MFA components by a factor of 20 (ablation) | MFA · CelebA | sharpness: -5.4 -> -4.0 | Improved |
| Increase the number of MFA components by a factor of 20 (ablation) | MFA · CelebA | overfitting | Hurt |
| Use WGAN-GP rather than WGAN on CelebA | GAN training · CelebA | NDB/K: 0.62 -> 0.27 (K=300) | Improved |
Closest papers by meaning
- No Modes Left Behind: Capturing the Data Distribution Effectively Using GANs (AAAI 2018 · similarity 0.87)
- PacGAN: The power of two samples in generative adversarial networks (NeurIPS 2018 · similarity 0.86)
- Unsupervised Image Generation With Infinite Generative Adversarial Networks (ICCV 2021 · similarity 0.86)
- Do GANs learn the distribution? Some Theory and Empirics (ICLR 2018 · similarity 0.85)
- Seeing What a GAN Cannot Generate (ICCV 2019 · similarity 0.85)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.83)
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.
