Research map / Label-efficient and robust learning / OOD and anomaly detection
Natural Synthetic Anomalies for Self-Supervised Anomaly Detection and Localization
Hannah M. Schlüter, Jeremy Tan, Benjamin Hou, Bernhard Kainz
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 NSA (Poisson blended, resized, shifted patches) instead of CutPaste patches | end-to-end anomaly segmentation model · MVTec AD image-level | AUROC: 84.5 -> 97.2 | Improved |
| use logistic labels instead of binary labels (ablation) | NSA · MVTec AD image-level | AUROC: 95.9 -> 97.2 | Improved |
| use bounded logistic labels instead of unbounded continuous labels (ablation) | NSA · MVTec AD image-level | AUROC: 95.2 -> 97.2 | Improved |
| use binary labels instead of unbounded continuous labels (ablation) | NSA · MVTec AD pixel-level | AUROC: 93.1 -> 96.3 | Improved |
| use logistic or binary labels instead of continuous (ablation) | NSA · rCXR (female patients) | AUROC: 93.0 -> 94.3 | No clear effect |
| use NSA instead of PII | anomaly self-supervision · rCXR (male patients) | AUROC: 91.7 -> 94.0 | Improved |
| shift and resize patches, as in NSA, instead of PII same-location blending | anomaly self-supervision · MVTec AD object classes image-level | AUROC: 87.7 -> 96.5 | Improved |
| add random number of patches instead of single patch (ablation) | NSA (logistic) · MVTec AD image-level | AUROC: 95.3 -> 97.2 | Improved |
| use NSA patch selection instead of CutPaste-style patch selection (ablation) | NSA (logistic) · MVTec AD image-level | AUROC: 91.4 -> 97.2 | Improved |
| mask patches with random ellipses (ablation) | NSA (logistic) · MVTec AD image-level | AUROC: 97.2 -> 97.0 | No clear effect |
| add foreground constraints (ablation) | NSA (logistic) · MVTec AD capsule image-level | AUROC: 91.7 -> 95.2 | Improved |
| use NSA instead of DRAEM without its reconstructive sub-network | anomaly localization model · MVTec AD image-level | AUROC: 93.9 -> 97.2 | Improved |
Closest papers by meaning
- Few-Shot Anomaly-Driven Generation for Anomaly Classification and Segmentation (ECCV 2024 · similarity 0.89)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.87)
- AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · similarity 0.87)
- RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection (CVPR 2024 · similarity 0.87)
- Exploring Intrinsic Normal Prototypes within a Single Image for Universal Anomaly Detection (CVPR 2025 · similarity 0.86)
- ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining (NeurIPS 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.
