Research map / Label-efficient and robust learning / OOD and anomaly detection
UniMMAD: Unified Multi-Modal and Multi-Class Anomaly Detection via MoE-Driven Feature Decompression
Yuan Zhao, Youwei Pang, Lihe Zhang, Hanqi Liu, Jiaming Zuo, Huchuan Lu, Xiaoqi Zhao
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 OCBE with the Feature Compression Module (ablation) | RD-based baseline · Mean over 5 datasets | AUC_I: 75.624 -> 77.367 | Improved |
| add the general-to-specific paradigm with cross attention (ablation) | baseline + FCM · Mean over 5 datasets | AUC_I: 77.367 -> 84.306 | Improved |
| add C-MoE decompression (ablation) | general-to-specific model · Mean over 5 datasets | AUC_I: 84.306 -> 91.145 | Improved |
| use cross-condition routing instead of plain MoE (ablation) | C-MoE · Mean over 5 datasets | AUC_I: 85.126 -> 91.145 | Improved |
| add routed experts (ablation) | C-MoE · Mean over 5 datasets | AUC_I: 85.417 -> 91.145 | Improved |
| add multi-scale experts (ablation) | C-MoE · Mean over 5 datasets | AUC_I: 88.900 -> 91.145 | Improved |
| add fixed expert (ablation) | C-MoE · Mean over 5 datasets | AUC_I: 89.424 -> 91.145 | Improved |
| add multi-scale experts (ablation) | C-MoE · MulSen-AD | AUC_I: 81.960 -> 85.460 | Improved |
| use MoE-in-MoE instead of plain MoE (ablation) | C-MoE · parameter efficiency | parameters: 75% fewer | Improved |
| use UniMMAD instead of specialized CFM | CFM · MVTec-3D | AUC_I: 92.443 -> 92.527 | No clear effect |
| use UniMMAD instead of INP-Former in super multi-class setting | INP-Former · VisA | MF1_P: 44.4 -> 47.2 | Improved |
| use UniMMAD instead of INP-Former | INP-Former · MVTec-AD pixel-level | MF1_P: 60.7 -> 60.2 | Hurt |
Closest papers by meaning
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.89)
- CNC: Cross-modal Normality Constraint for Unsupervised Multi-class Anomaly Detection (AAAI 2025 · similarity 0.88)
- Unlocking the Potential of Reverse Distillation for Anomaly Detection (AAAI 2025 · similarity 0.87)
- ResAD: A Simple Framework for Class Generalizable Anomaly Detection (NeurIPS 2024 · similarity 0.87)
- A Comprehensive Augmentation Framework for Anomaly Detection (AAAI 2024 · similarity 0.87)
- Generating and Reweighting Dense Contrastive Patterns for Unsupervised Anomaly Detection (AAAI 2024 · similarity 0.87)
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.
