Research map / Label-efficient and robust learning / OOD and anomaly detection
SoftPatch: Unsupervised Anomaly Detection with Noisy Data
Jiang Xi, Jianlin Liu, Jinbao Wang, Qiang Nie, Kai WU, Yong Liu, Chengjie Wang, Feng Zheng
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 LOF noise discriminator before coreset selection (ablation) | SoftPatch · MVTecAD-noise-0.1 overlap | pixel AUROC: 0.693 -> 0.963 | Improved |
| add LOF noise discriminator before coreset selection (ablation) | SoftPatch · MVTecAD-noise-0.1 overlap | image AUROC: 0.685 -> 0.984 | Improved |
| add soft weight re-weighting to LOF (ablation) | SoftPatch · MVTecAD-noise-0.1 overlap | pixel AUROC: 0.963 -> 0.969 | Improved |
| add soft weight re-weighting to LOF (ablation) | SoftPatch · MVTecAD-noise-0.1 overlap | image AUROC: 0.984 -> 0.982 | No clear effect |
| add soft weight to nearest discriminator (ablation) | SoftPatch · MVTecAD-noise-0.1 overlap | pixel AUROC: 0.944 -> 0.958 | Improved |
| add soft weight to Gaussian discriminator (ablation) | SoftPatch · MVTecAD-noise-0.1 no overlap | image AUROC: 0.927 -> 0.922 | No clear effect |
| use LOF instead of nearest/Gaussian discriminator (ablation) | SoftPatch · MVTecAD-noise-0.1 overlap | image AUROC: 0.966/0.925 -> 0.984 | Improved |
| use LOF-K >= 6 instead of small K (3) (ablation) | SoftPatch · MVTecAD overlap | pixel AUROC: 0.955 -> 0.975 | Improved |
| increase removal threshold tau toward 0.15 (ablation) | SoftPatch · MVTecAD overlap | image/pixel AUROC | Improved |
| use SoftPatch-LOF instead of PatchCore under noise | SoftPatch · MVTecAD-noise-0.1 overlap | pixel AUROC: 0.654 -> 0.969 | Improved |
| use SoftPatch instead of PatchCore with no added noise | SoftPatch · BTAD mean | image AUROC: 0.957 -> 0.977 | Improved |
| use random instead of greedy coreset sampling in PatchCore | SoftPatch · MVTecAD-noise-0.1 no overlap | image AUROC: 0.984 -> 0.896 | Hurt |
Later papers that compared with this method
4 later accepted papers report SoftPatch as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- BAAF: Universal Transformation of One-Class Classifiers for Unsupervised Image Anomaly Detection (ECCV 2026 · better on 4 of 4 shared results)
- Memory-Distilled Selection for Noise-Robust Anomaly Detection (ICML 2026 · better on 3 of 3 shared results)
- TailedCore: Few-Shot Sampling for Unsupervised Long-Tail Noisy Anomaly Detection (CVPR 2025 · better on 2 of 2 shared results)
- M3DM-NR: RGB-3D Noisy-Resistant Industrial Anomaly Detection via Multimodal Denoising (TPAMI 2025 · better on 1 of 1 shared result)
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
- TailedCore: Few-Shot Sampling for Unsupervised Long-Tail Noisy Anomaly Detection (CVPR 2025 · similarity 0.87)
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.86)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.86)
- CostFilter-AD: Enhancing Anomaly Detection through Matching Cost Filtering (ICML 2025 · similarity 0.86)
- Long-Tailed Anomaly Detection with Learnable Class Names (CVPR 2024 · similarity 0.86)
- AnomalyNCD: Towards Novel Anomaly Class Discovery in Industrial Scenarios (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.
