Research map / Label-efficient and robust learning / OOD and anomaly detection
CMDS-AD: Cross-Modal Dual-Stream Decoupling for Few-Shot Anomaly Detection
Junhao Cai, Deyu Zeng, Junhao Pang, JunYu Chen, Qiwei Liang, Xiaopin ZHONG, Zongze Wu
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 estimated (diffusion-normal) stream to real stream (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | AUPRO@1%: 0.400 -> 0.410 | Improved |
| add estimated stream to real stream (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | I-AUROC: 0.873 -> 0.871 | No clear effect |
| add real stream to estimated stream (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | I-AUROC: 0.813 -> 0.871 | Improved |
| use Feature Mapper instead of standard MLP fusion (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | AUPRO@1%: 0.399 -> 0.410 | Improved |
| use dual-stream architecture instead of MLP under identical diffusion priors (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | PRO@1%: 39.9 -> 41.0 | Improved |
| use multiplicative fusion instead of additive fusion of anomaly maps (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | AUPRO@1%: 0.401 -> 0.410 | Improved |
| use multiplicative fusion instead of additive fusion (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | I-AUROC: 0.875 -> 0.871 | Mixed |
| fuse 2D and 3D maps by multiplication instead of 2D map only (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | I-AUROC: 0.807 -> 0.871 | Improved |
| raise estimated-stream weight from 0.0 to 0.1 (ablation) | CMDS-AD · EyeCandies 2-shot | I-AUROC: 78.46% -> 80.20% | Improved |
| raise estimated-stream weight from 0.0 to 0.1 (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | AUPRO@30%: 95.63% -> 95.83% | Improved |
| raise estimated-stream weight to 1.0 (ablation) | CMDS-AD · EyeCandies 1-shot | I-AUROC: 75.85% | Hurt |
| vary synthetic samples N per real image from 1 to 3 (ablation) | CMDS-AD · MVTec 3D-AD 4-shot | PRO@1%: 40.4 (N=1), 40.6 (N=3) vs 41.0 | No clear effect |
Closest papers by meaning
- Align3D-AD: Cross-Modal Feature Alignment and Dual-Prompt Learning for Zero-shot 3D Anomaly Detection (NeurIPS 2026 · similarity 0.90)
- Beyond Single-Modal Boundary: Cross-Modal Anomaly Detection through Visual Prototype and Harmonization (CVPR 2025 · similarity 0.89)
- Rethinking Reverse Distillation for Multi-Modal Anomaly Detection (AAAI 2024 · similarity 0.88)
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.86)
- DZAD: Diffusion-based Zero-shot Anomaly Detection (AAAI 2025 · similarity 0.86)
- CNC: Cross-modal Normality Constraint for Unsupervised Multi-class Anomaly Detection (AAAI 2025 · 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.
