Research map / Label-efficient and robust learning / OOD and anomaly detection
Hierarchical Vector Quantized Transformer for Multi-class Unsupervised Anomaly Detection
Ruiying Lu, YuJie Wu, Long Tian, Dongsheng Wang, Bo Chen, Xiyang Liu, Ruimin Hu
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 |
|---|---|---|---|
| replace continuous latent features with vector-quantized normal prototypes (ablation) | HVQ-Trans on MVTec-AD · MVTec-AD | image-level detection AUROC: 70.5 to 98.0 | Improved |
| use hierarchical VQLayers rather than a single VQLayer (ablation) | HVQ-Trans on MVTec-AD · MVTec-AD | image-level detection AUROC: 96.2 to 97.1 | Improved |
| add codebook switching (ablation) | HVQ-Trans with hierarchical VQLayers · MVTec-AD | image-level detection AUROC: 97.1 to 97.2 | Improved |
| add POT-based scoring (ablation) | HVQ-Trans with hierarchical VQLayers · MVTec-AD | image-level detection AUROC: 97.1 to 97.4 | Improved |
| add codebook and expert switching (ablation) | HVQ-Trans with hierarchical VQLayers · MVTec-AD | image-level detection AUROC: 97.1 to 97.6 | Improved |
| add POT to the model with codebook and expert switching (ablation) | HVQ-Trans · MVTec-AD | image-level detection AUROC: 97.6 to 98.0 | Improved |
| concatenate global prototypes with earlier-layer features for quantization (ablation) | HVQ-Trans hierarchy · MVTec-AD | image-level detection AUROC: 97.2 to 98.0 | Improved |
| use 512 prototypes rather than 1024 or 256 (ablation) | HVQ-Trans · MVTec-AD | image-level detection AUROC: 1024: 97.1; 512: 98.0; 256: 97.0 | Mixed |
| use HVQ-Trans instead of UniAD | one-for-all anomaly detection · MVTec-AD, mean image-level | AUROC: 96.5 to 98.0 | Improved |
| use HVQ-Trans instead of UniAD | one-for-all anomaly localization · MVTec-AD, mean pixel-wise | AUROC: 96.8 to 97.3 | Improved |
| use HVQ-Trans instead of UniAD | one-for-all anomaly detection · VisA, mean image-level | AUROC: 91.9 to 93.2 | Improved |
| use HVQ-Trans instead of UniAD | one-for-all anomaly localization · VisA, mean pixel-wise | AUROC: 98.6 to 98.7 | Improved |
Closest papers by meaning
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.88)
- A Comprehensive Augmentation Framework for Anomaly Detection (AAAI 2024 · similarity 0.86)
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.86)
- CNC: Cross-modal Normality Constraint for Unsupervised Multi-class Anomaly Detection (AAAI 2025 · similarity 0.86)
- UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection (CVPR 2025 · similarity 0.86)
- Long-Tailed Anomaly Detection with Learnable Class Names (CVPR 2024 · 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.
