Research map / Label-efficient and robust learning / OOD and anomaly detection
ReFP-AD: Rectified Flow Preconditioning for Energy-Based Anomaly Detection
Camile Lendering, Erkut Akdag, Joaquin Figueira Chacon, Egor Bondarev
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 |
|---|---|---|---|
| add OT-coupled rectified flow preconditioning (ablation) | ReFP-AD · VisA (unified) | Image AUROC: 87.0 -> 97.3 | Improved |
| add OT-coupled rectified flow preconditioning (ablation) | ReFP-AD · MVTec-AD (unified) | Image AUROC: 91.1 -> 98.6 | Improved |
| use preconditioned SGLD instead of standard SGLD (ablation) | ReFP-AD · VisA (unified) | Image AUROC: 75.3 -> 97.3 | Improved |
| increase pSGLD steps from 10 to 20 (ablation) | ReFP-AD · MVTec-AD (unified) | Image AUROC: 54.7 -> 98.2 | Improved |
| increase pSGLD steps from 40 to 80 (ablation) | ReFP-AD · MVTec-AD (unified) | Image AUROC: 98.5 -> 98.6 | No clear effect |
| train per-category models instead of one unified model (ablation) | ReFP-AD · MVTec-AD (per-category vs unified) | Image AUROC: 98.6 -> 99.2 | No clear effect |
| train per-category models instead of one unified model (ablation) | ReFP-AD · VisA (per-category vs unified) | Image AUROC: 97.3 -> 97.7 | No clear effect |
| use larger DINOv2 backbone (ViT-B to ViT-G) (ablation) | ReFP-AD · VisA (unified) | Image AUROC: 95.7 -> 97.3 | Improved |
| replace DINOv2-G with DINOv3-7B (4096-D) (ablation) | ReFP-AD · VisA (unified) | Image AUROC: 97.3 -> 96.8 | No clear effect |
| score with EBM gradient norm instead of flow magnitude (ablation) | ReFP-AD · MVTec-AD (unified) | Image AUROC: 97.51 -> 98.60 | Improved |
| select flow checkpoint by SGLD-fitness instead of minimum flow loss (ablation) | ReFP-AD · VisA (unified) | Image AUROC: 73.05 -> 97.3 | Improved |
| raise input resolution from 224 to 448 (ablation) | ReFP-AD · MVTec-AD (unified) | Image AUROC: 98.2 -> 98.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Dinomaly — Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025) · better on 0 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Energy-Based Models for Anomaly Detection: A Manifold Diffusion Recovery Approach (NeurIPS 2023 · similarity 0.87)
- Scalable, Explainable and Provably Robust Anomaly Detection with One-Step Flow Matching (NeurIPS 2025 · similarity 0.86)
- Balanced Conic Rectified Flow (NeurIPS 2025 · similarity 0.85)
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.85)
- One-for-More: Continual Diffusion Model for Anomaly Detection (CVPR 2025 · similarity 0.85)
- A Unified Latent Schrodinger Bridge Diffusion Model for Unsupervised Anomaly Detection and Localization (CVPR 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.
