Research map / Probabilistic and scientific modeling / Variational and generative sampling
Probabilistic Precision and Recall Towards Reliable Evaluation of Generative Models
Dogyun Park, Suhyun Kim
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 probabilistic scoring rule (PSR) instead of binary scoring rule for precision | P-precision · Gaussian toy with outlier | robustness to outliers: I-precision +0.2 vs P-precision +0.05 under 5% outliers (ADM, ImageNet 128) | Improved |
| remove probabilistic subsupport estimation (A) from PSR (ablation) | P-precision · Gaussian toy with outlier, u in [-3,-1] | outlier robustness | Improved |
| use identical radius R for all samples instead of sample-specific kNN radius (B) (ablation) | P-precision · Gaussian toy with outlier | outlier robustness and consistency over k | Improved |
| use symmetric scoring for diversity (P-recall) instead of Coverage | diversity metric · N(0,I) vs N(0,vI), v in [0.2,1.5] | sensitivity to diversity change: Coverage stays at 1 for v 0.2 to 1 | Improved |
| use PP&PR instead of IP&IR for small sample size | precision/recall metrics · identical Gaussians, varying N | bias and standard deviation | Improved |
| Density metric with small N | Density · identical Gaussians, N <= 10000 | bias and standard deviation | Hurt |
| increase classifier guidance scale from 0.5 to 10.0 in ADM | ADM · ImageNet 128 | fidelity-diversity trade-off captured by metrics | Mixed |
| F1 of PP&PR compared with FID ranking | PP&PR · ImageNet, AFHQv2, LSUN Bedroom | consistency with FID | Improved |
Closest papers by meaning
- A Unifying Information-theoretic Perspective on Evaluating Generative Models (AAAI 2025 · similarity 0.91)
- Efficient Precision and Recall Metrics for Assessing Generative Models using Hubness-aware Sampling (ICML 2024 · similarity 0.89)
- PQMass: Probabilistic Assessment of the Quality of Generative Models using Probability Mass Estimation (ICLR 2025 · similarity 0.87)
- Position: All Current Generative Fidelity and Diversity Metrics are Flawed (ICML 2025 · similarity 0.87)
- Why Knowledge Distillation Works in Generative Models: A Minimal Working Explanation (NeurIPS 2025 · similarity 0.86)
- Probably Approximately Precision and Recall Learning (NeurIPS 2025 · similarity 0.84)
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.
