Research map / Label-efficient and robust learning / OOD and anomaly detection
Locally Varying Distance Transform for Unsupervised Visual Anomaly Detection
Wen-Yan Lin, Zhonghang Liu, Siying Liu
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 instance (layer) normalization to the features (ablation) | LVAD and other anomaly detectors · STL-10 | normality probability separation | Improved |
| increase the number of clusters (local projections) (ablation) | LVAD · STL-10 and MNIST | average AUROC | Improved |
| use ViT features instead of ResNet-50 (ablation) | LVAD · STL-10 (20% anomalies) | average AUROC: 0.983 -> 0.922 | Hurt |
| use SWIN-B features instead of ResNet-50 (ablation) | LVAD · STL-10 (20% anomalies) | average AUROC: 0.983 -> 0.963 | Hurt |
| use SWIN-B features instead of ResNet-50 (ablation) | LVAD · MIT-Places-Small (20% anomalies) | average AUROC: 0.891 -> 0.899 | Improved |
| use ViT features instead of ResNet-50 (ablation) | LVAD · MIT-Places-Small (20% anomalies) | average AUROC: 0.891 -> 0.870 | Hurt |
| move from single-normal to multi-normal class setting | LVAD · STL-10 | average AUROC: 0.986 -> 0.909 | Hurt |
| compare LVAD to OC-SVM in multi-normal detection | LVAD · STL-10 multi-normal | average AUROC: 0.778 -> 0.909 | Improved |
| increase anomaly percentage from 0.1% to 30% (performance reversal) | Shell-Renormalization · STL-10 | AUROC: 0.803 -> 0.999 | Mixed |
| increase anomaly percentage from 0.1% to 30% | OC-SVM · STL-10 | AUROC: 0.996 -> 0.777 | Hurt |
| increase anomaly percentage from 0.1% to 30% | LVAD · STL-10 | AUROC: 0.998 -> 0.977 | No clear effect |
| PCA dimension reduction to very low dimension in naive KDE detector (ablation) | naive KDE · STL-10 and MNIST | AUROC | Mixed |
Closest papers by meaning
- UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection (CVPR 2025 · similarity 0.88)
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.87)
- Odd-One-Out: Anomaly Detection by Comparing with Neighbors (CVPR 2025 · similarity 0.86)
- CostFilter-AD: Enhancing Anomaly Detection through Matching Cost Filtering (ICML 2025 · similarity 0.86)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.86)
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (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.
