Research map / Image generation / GANs
Towards GAN Benchmarks Which Require Generalization
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 |
|---|---|---|---|
| use CNN divergence instead of Inception Score to assess memorization | Image generation evaluation · 32×32 ImageNet | n to Win: 32 -> >1M | Helped |
| use CNN divergence instead of FID to assess memorization | Image generation evaluation · 32×32 ImageNet | n to Win: 1024 -> >1M | Helped |
| reduce critic channels to 1/4 at each layer (ablation) | CNN divergence · 32×32 ImageNet | n to Win: >1M -> 32,768 | Hurt |
| train with WGAN-GP rather than memorize the training set | CIFAR-10 image generation · CIFAR-10 | CNN divergence: 14.62 -> 12.97 | Helped |
| train a GAN rather than PixelCNN++ | CIFAR-10 image generation · CIFAR-10 | CNN divergence: 16.17 -> 12.97 | Helped |
| train a GAN rather than an IAF VAE | CIFAR-10 image generation · CIFAR-10 | CNN divergence: 18.11 -> 12.97 | Helped |
| estimate CNN divergence with a small rather than large test set | CNN divergence evaluation · 32×32 ImageNet | model ranking correlation | No clear effect |
| evaluate WGAN-GP samples rather than training-set memorization with Inception Score or FID | CIFAR-10 image generation · CIFAR-10 | evaluation score | Hurt |
Closest papers by meaning
- Quantitatively Evaluating GANs With Divergences Proposed for Training (ICLR 2018 · similarity 0.92)
- A Domain Agnostic Measure for Monitoring and Evaluating GANs (NeurIPS 2019 · similarity 0.89)
- Precision-Recall Divergence Optimization for Generative Modeling with GANs and Normalizing Flows (NeurIPS 2023 · similarity 0.89)
- Assessing Generative Models via Precision and Recall (NeurIPS 2018 · similarity 0.88)
- Improved Precision and Recall Metric for Assessing Generative Models (NeurIPS 2019 · similarity 0.88)
- SMaRt: Improving GANs with Score Matching Regularity (ICML 2024 · 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.