Research map / Image generation / Autoregressive and latent generation
Projection Regret: Reducing Background Bias for Novelty Detection via Diffusion Models
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 |
|---|---|---|---|
| add Projection Regret score (regret between single-step and full-step projection) over projection-only LPIPS (ablation) | Projection Regret · CIFAR-10 vs CIFAR-100 | AUROC: 0.667 -> 0.760 | Helped |
| add time ensemble over alpha and beta (ablation) | Projection Regret · CIFAR-10 vs CIFAR-100 | AUROC: 0.760 -> 0.775 | Helped |
| use projection-only LPIPS score instead of LMD (ablation) | consistency-model OOD detector · CIFAR-10 vs SVHN | AUROC: 0.979 -> 0.966 | Hurt |
| use projection-only LPIPS score instead of LMD (ablation) | consistency-model OOD detector · CIFAR-10 vs ImageNet | AUROC: 0.686 -> 0.718 | Helped |
| replace LPIPS distance with SSIM (ablation) | Projection Regret · CIFAR-10 vs SVHN | AUROC: 0.993 -> 0.328 | Hurt |
| replace LPIPS distance with UNet decoder feature distance (ablation) | Projection Regret · CIFAR-10 vs LSUN | AUROC: 0.837 -> 0.865 | Helped |
| replace LPIPS distance with UNet decoder feature distance (ablation) | Projection Regret · CIFAR-10 vs SVHN | AUROC: 0.993 -> 0.917 | Hurt |
| use EDM with Heun 2nd-order sampler instead of consistency model for full-step projection (ablation) | Projection Regret · CIFAR-10 vs CIFAR-100 | AUROC: 0.761 -> 0.763 | No clear effect |
| increase ensemble size from n_alpha=10, n_beta=5 to n_alpha=40, n_beta=10 (ablation) | Projection Regret · CIFAR-10 vs ImageNet | AUROC: 0.797 -> 0.814 | Helped |
| combine Projection Regret with Rot+Trans by multiplying scores | Rot+Trans · CIFAR-10 one-class mean | AUROC: 0.886 -> 0.908 | Helped |
| use Projection Regret instead of Rot+Trans on airplane class | one-class classifier · CIFAR-10 one-class Plane | AUROC: 0.760 -> 0.864 | Helped |
| use Projection Regret instead of MSMA/LMD with consistency models | OOD detector · CIFAR-100 vs SVHN | AUROC: 0.968 -> 0.945 | Hurt |
Closest papers by meaning
- Universal Novelty Detection Through Adaptive Contrastive Learning (CVPR 2024 · similarity 0.87)
- One-for-More: Continual Diffusion Model for Anomaly Detection (CVPR 2025 · similarity 0.87)
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.86)
- Distribution Prototype Diffusion Learning for Open-set Supervised Anomaly Detection (CVPR 2025 · similarity 0.86)
- An Interpretable Evaluation of Entropy-based Novelty of Generative Models (ICML 2024 · similarity 0.86)
- Diffusion-based Semantic Outlier Generation via Nuisance Awareness for Out-of-Distribution Detection (AAAI 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.