Research map / Label-efficient and robust learning / OOD and anomaly detection
GPFlow: Gaussian Prototype Probability Flow for Unsupervised Multi-Modal Anomaly Detection
Yiting Li, Xulei Yang, Jingyi Liao, Jing Zhang, Fayao 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 covariance-aware PMP router instead of attention-based prototype router (ablation) | GPFlow · MVTec-3D-AD (10-shot) | I-AUROC: 0.883 -> 0.935 | Improved |
| add cross-modal reconstruction to intra-modal (coupled design) (ablation) | GPFlow · MVTec-3D-AD (10-shot) | I-AUROC: 0.915 -> 0.935 | Improved |
| add instance-aware prior calibration (ablation) | GPFlow · MVTec-3D-AD (10-shot) | I-AUROC: 0.925 -> 0.935 | Improved |
| use Gaussian prototypes with PMP instead of VQ codebook with soft assignment (ablation) | GPFlow · MVTec-3D-AD (10-shot) | I-AUROC: 0.885 -> 0.935 | Improved |
| use VQ codebook with optimal transport instead of Gaussian + PMP (ablation) | GPFlow · MVTec-3D-AD (10-shot) | I-AUROC: 0.935 -> 0.751 | Hurt |
| use analytical PMP router instead of learned Schrodinger Bridge (DPDL) (ablation) | GPFlow · MVTec-3D-AD (10-shot) | I-AUROC: 0.908 -> 0.935 | Improved |
| use cross-modal reconstruction only instead of intra-modal only (ablation) | GPFlow · MVTec-3D-AD (10-shot) | I-AUROC: 0.915 -> 0.775 | Hurt |
| increase MSM selection ratio rho beyond 0.6 to 0.8 (ablation) | GPFlow IPC · MVTec-3D-AD | I-AUROC: 0.911 -> 0.907 | No clear effect |
| inject 100% anomalous patches into prototype calibration (ablation) | GPFlow IPC · MVTec-3D-AD | I-AUROC: 0.894 -> 0.872 | Hurt |
| inject up to 50% anomalous patches into prototype calibration (ablation) | GPFlow IPC · MVTec-3D-AD | I-AUROC: 0.894 -> 0.895 | Mixed |
| train with all normal data instead of few-shot | GPFlow vs FIND · MVTec-3D-AD (all-shot) | I-AUROC: 0.978 -> 0.970 | Hurt |
| use GPFlow instead of FIND in compute cost | GPFlow · Efficiency | FLOPs (G): 728.46 -> 129.27 | Improved |
Closest papers by meaning
- Beyond Single-Modal Boundary: Cross-Modal Anomaly Detection through Visual Prototype and Harmonization (CVPR 2025 · similarity 0.86)
- CNC: Cross-modal Normality Constraint for Unsupervised Multi-class Anomaly Detection (AAAI 2025 · similarity 0.86)
- Rethinking Reverse Distillation for Multi-Modal Anomaly Detection (AAAI 2024 · similarity 0.85)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.85)
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.85)
- Scalable, Explainable and Provably Robust Anomaly Detection with One-Step Flow Matching (NeurIPS 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.
