Research map / Label-efficient and robust learning / OOD and anomaly detection
SANFlow: Semantic-Aware Normalizing Flow for Anomaly Detection
Daehyun Kim, Sungyong Baik, Tae Hyun Kim
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 |
|---|---|---|---|
| train with all three losses (add L_BCE and L_KL) vs plain NLL (ablation) | NF baseline · STC image-wise | AUROC: 72.6 -> 76.1 | Improved |
| train with all three losses vs plain NLL (ablation) | NF baseline · MVTec image-wise | AUROC: 98.3 -> 98.7 | Improved |
| add L_KL on top of Model (1d) (ablation) | SANFlow · STC image-wise | AUROC: 74.2 -> 76.1 | Improved |
| add L_BCE on top of Model (1c) (ablation) | SANFlow · STC image-wise | AUROC: 73.1 -> 76.1 | Improved |
| add L_KL only (Model 1c) to Model (1b) (ablation) | NF with NLL Eq.3&5 · MVTec image-wise | AUROC: 98.2 -> 98.3 | No clear effect |
| add synthetic anomaly augmentation (ablation) | SANFlow · STC image-wise | AUROC: 73.6 -> 74.2 | Improved |
| add synthetic anomaly augmentation (ablation) | SANFlow · MVTec image-wise | AUROC: 98.4 -> 98.5 | Improved |
| estimate variance with network, mean fixed (ablation) | SANFlow · STC image-wise | AUROC: 75.9 -> 76.1 | Improved |
| estimate variance with network, mean fixed (ablation) | SANFlow · MVTec image-wise | AUROC: 97.6 -> 98.7 | Improved |
| also estimate mean (ablation) | SANFlow · STC image-wise | AUROC: 76.1 -> 74.5 | Hurt |
| mean margin 1.0 vs 0.5 vs 1.5 (ablation) | SANFlow · MVTec image-wise | AUROC: 98.6, 98.7, 98.1 | Mixed |
| add statistics prediction (ablation) | CFLOW-AD · STC image-wise | AUROC: 72.63 -> 73.6 | Improved |
Closest papers by meaning
- Mixture Prototype Flow Matching for Open-Set Supervised Anomaly Detection (ICML 2026 · similarity 0.86)
- Normalizing Flows are Capable Generative Models (ICML 2025 · similarity 0.86)
- Understanding Likelihood of Normalizing Flow and Image Complexity through the Lens of Out-of-Distribution Detection (AAAI 2024 · similarity 0.85)
- Language-Assisted Feature Transformation for Anomaly Detection (ICLR 2025 · similarity 0.84)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.84)
- CNC: Cross-modal Normality Constraint for Unsupervised Multi-class Anomaly Detection (AAAI 2025 · 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.
