Research map / Label-efficient and robust learning / OOD and anomaly detection
DecAD: Decoupling Anomalies in Latent Space for Multi-Class Unsupervised Anomaly Detection
Xiaolei Wang, Xiaoyang Wang, Huihui Bai, Eng Gee Lim, Jimin Xiao
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 anomaly synthesis (adversarial perturbation) to the distillation baseline (ablation) | VDF · MVTec AD | image AUROC: 97.0 -> 97.9 | Improved |
| add anomaly decoupling process on top of anomaly synthesis (ablation) | VDF + AS · MVTec AD | pixel AP: 55.1 -> 60.7 | Improved |
| add both synthesis and decoupling to the baseline (ablation) | VDF · MVTec AD | image AUROC: 97.0 -> 99.0 | Improved |
| use general INN (Glow) instead of MLP as decoupling network (ablation) | DecAD · MVTec AD | mean of seven metrics: 84.7 -> 85.4 | Improved |
| use globally bounded INN instead of general INN (ablation) | DecAD · MVTec AD | mean of seven metrics: 86.1 -> 87.2 | Improved |
| increase expansion ratio from 1.5 to 2.0 (ablation) | DecAD bounded INN · MVTec AD | mean of seven metrics: 87.2 -> 86.0 | Hurt |
| attack distillation loss instead of no adversarial noise (ablation) | DecAD anomaly synthesis · MVTec AD | image AUROC: 98.3 -> 86.2 | Hurt |
| attack decoupling loss instead of no adversarial noise (ablation) | DecAD anomaly synthesis · MVTec AD | image AUROC: 98.3 -> 74.3 | Hurt |
| attack classification loss instead of no adversarial noise (ablation) | DecAD anomaly synthesis · MVTec AD | image AUROC: 98.3 -> 98.6 | Improved |
| attack entropy of classification logits instead of classification loss (ablation) | DecAD anomaly synthesis · MVTec AD | image AUROC: 98.6 -> 99.0 | Improved |
| use uniform U[-1,1] instead of Gaussian N(0,1) initialization with entropy attack (ablation) | DecAD anomaly synthesis · MVTec AD | pixel AUROC: 97.5 -> 97.8 | Improved |
| use DINOv2 ViT-B/14 instead of ViT-B/16 encoder (ablation) | DecAD · MVTec AD | mean of seven metrics: 86.7 -> 87.2 | Improved |
Closest papers by meaning
- ResAD: A Simple Framework for Class Generalizable Anomaly Detection (NeurIPS 2024 · similarity 0.88)
- Unlocking the Potential of Reverse Distillation for Anomaly Detection (AAAI 2025 · similarity 0.88)
- Long-Tailed Anomaly Detection with Learnable Class Names (CVPR 2024 · similarity 0.88)
- CNC: Cross-modal Normality Constraint for Unsupervised Multi-class Anomaly Detection (AAAI 2025 · similarity 0.87)
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.87)
- Dinomaly: The Less Is More Philosophy in Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · 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.
