Research map / Image generation / Autoregressive and latent generation
Rarity Score : A New Metric to Evaluate the Uncommonness of Synthesized Images
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 |
|---|---|---|---|
| Measure generated images by the minimum radius of real-data k-NN spheres containing each image | Synthesized images · FFHQ, AFHQ-Wild, MetFaces | rarity score | Helped |
| Increase truncation parameter ψ | StyleGAN2 · FFHQ | rarity score | Helped |
| Use truncation parameter ψ ≤ 0.2 instead of larger ψ | StyleGAN2 · FFHQ | rarity score distribution | Hurt |
| Use StyleGAN2 rather than the other compared generators | Generative models trained on FFHQ · FFHQ | mean rarity score of top 0.1% samples | Helped |
| Use UnetGAN rather than the other compared generators | Generative models trained on FFHQ · FFHQ | mean rarity score | Hurt |
| Compare datasets on the union of their real-data manifolds | CelebA-HQ and FFHQ · Human-face datasets | rarity score distribution | Helped |
| Use FFHQ rather than CelebA-HQ | Human-face datasets · Human-face datasets | rarity score | Helped |
| Use VGG16 rather than DINO or CLIP as the feature extractor | Rarity scoring of StyleGAN2 generations · FFHQ | identified image characteristics | Mixed |
| Set k to 3 for the neighborhood size | Rarity score · Real-manifold samples | Spearman rank correlation | Helped |
Closest papers by meaning
- Image Generation Diversity Issues and How to Tame Them (CVPR 2025 · similarity 0.90)
- Anomaly Score: Evaluating Generative Models and Individual Generated Images based on Complexity and Vulnerability (CVPR 2024 · similarity 0.89)
- Attribute Based Interpretable Evaluation Metrics for Generative Models (ICML 2024 · similarity 0.86)
- Efficient Precision and Recall Metrics for Assessing Generative Models using Hubness-aware Sampling (ICML 2024 · similarity 0.85)
- RA-ClipScore: Making Generative Model Evaluation More Interpretable (NeurIPS 2026 · similarity 0.85)
- Position: All Current Generative Fidelity and Diversity Metrics are Flawed (ICML 2025 · similarity 0.85)
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.