Research map / Image generation / Fast diffusion sampling
GEPC: Group-Equivariant Posterior Consistency for Out-of-Distribution Detection in Diffusion Models
Rouzoumka Yadang Alexis, Jean Pinsolle, Eugénie TERREAUX, Christèle Morisseau, Jean-Philippe Ovarlez, Chengfang Ren
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 GEPC instead of SCOPED | CelebA-32 diffusion backbone · Average over 9 ID/OOD pairs at 32x32 | AUROC: 0.892 -> 0.908 | Improved |
| use GEPC instead of DiffPath | CelebA-32 diffusion backbone · Average over 9 ID/OOD pairs at 32x32 | AUROC: 0.918 -> 0.908 | Hurt |
| use GEPC instead of MSMA | CelebA-32 diffusion backbone · Average over 9 ID/OOD pairs at 32x32 | AUROC: 0.928 -> 0.908 | Hurt |
| use GEPC instead of LMD | CelebA-32 diffusion backbone · Average over 9 ID/OOD pairs at 32x32 | AUROC: 0.868 -> 0.908 | Improved |
| use GEPC instead of DiffPath | CelebA-32 diffusion backbone · CelebA vs CIFAR-10 | AUROC: 0.998 -> 1.00 | Improved |
| use GEPC instead of SCOPED | CelebA-32 diffusion backbone · CIFAR-10 vs CelebA | AUROC: 0.940 -> 0.999 | Improved |
| use GEPC instead of SCOPED | CelebA-32 diffusion backbone · SVHN vs CIFAR-10 | AUROC: 0.971 -> 0.880 | Hurt |
| use GEPC instead of DiffPath | CelebA-32 diffusion backbone · CIFAR-10 vs CIFAR-100 | AUROC: 0.590 -> 0.554 | Hurt |
| use GEPC instead of DiffPath for compute | CelebA-32 diffusion backbone · Compute F+J | forward-equivalent evaluations: 16F+0J vs 10F+0J | Improved |
| add ID-only CV-based timestep selection (ablation) | GEPC · 32x32 benchmarks | AUROC | Improved |
| use all group elements instead of a single transform (ablation) | GEPC · 32x32 benchmarks | AUROC | Improved |
Closest papers by meaning
- Backbone-Equated Diffusion OOD via Sparse Internal Snapshots (NeurIPS 2026 · similarity 0.87)
- Whitened Score Diffusion: A Structured Prior for Imaging Inverse Problems (NeurIPS 2025 · similarity 0.86)
- Diffusion Models With Learned Adaptive Noise (NeurIPS 2024 · similarity 0.86)
- Detecting Out-of-Distribution Through the Lens of Neural Collapse (CVPR 2025 · similarity 0.86)
- GOOD: Training-Free Guided Diffusion Sampling for Out-of-Distribution Detection (NeurIPS 2025 · similarity 0.86)
- Rethinking Diffusion Posterior Sampling: From Conditional Score Estimator to Maximizing a Posterior (ICLR 2025 · similarity 0.86)
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.
