Research map / Label-efficient and robust learning / OOD and anomaly detection
Continuous Memory Representation for Anomaly Detection
Joo Chan Lee, Taejune Kim, Eunbyung Park, Simon S Woo, Jong Hwan Ko
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 attention memory with continuous grid memory (local) (ablation) | CRAD · MVTec AD unified detection | AUROC: 95.9 -> 98.6 | Improved |
| enlarge attention memory from 64 to 256 entries (ablation) | attention memory baseline · MVTec AD unified detection | AUROC: 95.9 -> 93.9 | Hurt |
| enlarge VQ memory from 64 to 256 entries (ablation) | VQ memory baseline · MVTec AD unified detection | AUROC: 96.9 -> 97.8 | Improved |
| use continuous grid instead of VQ memory (global) (ablation) | CRAD global representation · MVTec AD unified detection | AUROC: 82.2 -> 92.3 | Improved |
| add global representation to local representation (ablation) | CRAD · MVTec AD unified detection | AUROC: 98.6 -> 98.8 | Improved |
| add feature refinement (ablation) | CRAD · MVTec AD unified detection | AUROC: 98.8 -> 99.1 | Improved |
| add coordinate jittering (ablation) | CRAD · MVTec AD unified detection | AUROC: 99.1 -> 99.3 | Improved |
| combine MSE and cosine similarity in refinement instead of either alone (ablation) | CRAD · MVTec AD unified detection | AUROC: 99.0 -> 99.3 | Improved |
| change grid dimension of local or global representation (ablation) | CRAD · MVTec AD unified | AUROC: 99.2 / 99.3 / 99.1 vs 99.3 | No clear effect |
| use CRAD-S, a downsized convolutional network (ablation) | CRAD · MVTec AD unified | detection AUROC: 99.3 -> 99.0 with 14.6M -> 4.5M params | Mixed |
| train in a unified multi-class setting rather than separate models | CRAD · MVTec AD detection | mean AUROC: 99.4 -> 99.3 | No clear effect |
| reduce training data to 1 normal image per class | CRAD · MVTec AD detection | AUROC: 83.3 at 1-shot vs 92.4 at 4-shot | Hurt |
Closest papers by meaning
- ResAD: A Simple Framework for Class Generalizable Anomaly Detection (NeurIPS 2024 · similarity 0.90)
- ADPretrain: Advancing Industrial Anomaly Detection via Anomaly Representation Pretraining (NeurIPS 2025 · similarity 0.86)
- Normal-Abnormal Guided Generalist Anomaly Detection (NeurIPS 2025 · similarity 0.86)
- AnomalyNCD: Towards Novel Anomaly Class Discovery in Industrial Scenarios (CVPR 2025 · similarity 0.86)
- A Diffusion-Based Framework for Multi-Class Anomaly Detection (AAAI 2024 · similarity 0.86)
- Unsupervised Continual Anomaly Detection with Contrastively-Learned Prompt (AAAI 2024 · 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.
