Research map / Label-efficient and robust learning / OOD and anomaly detection
Hierarchical Gaussian Mixture Normalizing Flow Modeling for Unified Anomaly Detection
Xincheng Yao, Ruoqi Li, Zefeng Qian, lu wang, Chongyang Zhang
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 a single latent Gaussian center with hierarchical Gaussian mixture modeling (ablation) | NF-based unified AD (CFLOW-style) · MVTecAD | image-level AUROC: 89.0 -> 97.7 | Improved |
| use fixed multiple class centers instead of a single center (ablation) | NF-based unified AD · MVTecAD | image-level AUROC: 89.0 -> 93.1 | Improved |
| add learnable inter-class Gaussian mixture modeling over fixed centers (ablation) | HGAD · MVTecAD | image-level AUROC: 93.1 -> 94.5 | Improved |
| add mutual information maximization loss (ablation) | HGAD · MVTecAD | image-level AUROC: 94.5 -> 96.3 (detection); 96.7 -> 96.6 (localization) | Mixed |
| add intra-class mixed class centers learning (ablation) | HGAD (FMC baseline) · MVTecAD | image-level AUROC: 93.1 -> 96.3 | Improved |
| combine log-likelihood with inter-class entropy as anomaly score (ablation) | HGAD · MVTecAD | image-level AUROC: 94.4 -> 97.7 | Improved |
| increase intra-class centers from 10 to 20 (ablation) | HGAD · MVTecAD | image-level AUROC: 97.7 -> 97.4 | No clear effect |
| increase mutual information loss weight lambda2 from 1 to 100 (ablation) | HGAD · MVTecAD | image-level AUROC: 96.5 -> 98.4 | Improved |
| use two-stage optimization for inter-class then intra-class centers (ablation) | HGAD · MVTecAD | image-level AUROC: 96.5 -> 97.7 | Improved |
| apply HGAD to the NF unified AD model instead of FastFlow | FastFlow · MVTecAD (unified) | image-level AUROC: 91.8 -> 98.4 | Improved |
| HGAD versus UniAD on MVTec3D-RGB | UniAD · MVTec3D-RGB (unified) | image-level AUROC: 77.5 -> 87.1 | Improved |
Closest papers by meaning
- Normal-Abnormal Guided Generalist Anomaly Detection (NeurIPS 2025 · similarity 0.87)
- ResAD: A Simple Framework for Class Generalizable Anomaly Detection (NeurIPS 2024 · similarity 0.86)
- Towards a Unified Framework of Clustering-based Anomaly Detection (ICML 2025 · similarity 0.85)
- ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining (NeurIPS 2025 · similarity 0.85)
- CNC: Cross-modal Normality Constraint for Unsupervised Multi-class Anomaly Detection (AAAI 2025 · similarity 0.85)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.84)
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.
