Research map / Label-efficient and robust learning / OOD and anomaly detection
Rethinking Continual Anomaly Detection on the Edge: Benchmarking Under Realistic Industrial Conditions
Chad Weatherly, Sen Lin
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 |
|---|---|---|---|
| use DINOSaur (frozen DINOv3 + spatial coreset memory) instead of dedicated CAD methods | UCAD/DNE/IUF · MVTec-AD | image AUROC: 0.798 -> 0.968 (vs UCAD) | Improved |
| DINOSaur vs PatchCore with replay | PatchCore · MVTec-AD | image AUROC: 0.925 -> 0.968 | Improved |
| DINOSaur vs PatchCore with replay | PatchCore · MVTec-LOCO | image AUROC: 0.743 -> 0.768 | Improved |
| dedicated CL mechanisms vs simple experience replay on traditional AD | DNE, IUF, UCAD · MVTec-AD | image AUROC: PatchCore+replay 0.925 vs UCAD 0.798 | Hurt |
| replay for EfficientAD | EfficientAD · MVTec-AD | image AUROC: 0.529 (near chance) | No clear effect |
| DINOSaur under geometric drift | DINOSaur · MTD-Geo | image AUROC: 0.519 (chance) | No clear effect |
| DINOSaur vs PatchCore on color drift | PatchCore · MTD-Color | image AUROC: 0.836 -> 0.928 | Improved |
| DINOSaur vs PatchCore latency on Jetson | PatchCore · Jetson Orin Nano | latency ms: 119.0 -> 91.6 | Improved |
| increase coreset ratio from 10% to 20% (ablation) | DINOSaur · Avg AUROC over 5 protocols | AUROC: 0.801 -> 0.816 | Improved |
| decrease coreset ratio from 10% to 1% (ablation) | DINOSaur · Avg AUROC over 5 protocols | AUROC: 0.801 -> 0.780 | Hurt |
| increase neighborhood radius from 0 to 3 at rho=10% (ablation) | DINOSaur · Avg AUROC over 5 protocols | AUROC: 0.787 -> 0.801 | Improved |
| pixel-level vs image-level evaluation | all methods · MVTec-AD, LOCO, MTD-color, MTD-blur | AUROC correlation: r = 0.945 | No clear effect |
Closest papers by meaning
- InCLAD: A Continual Learning Benchmark for Industrial Visual Anomaly Detection (NeurIPS 2026 · similarity 0.89)
- Unsupervised Continual Anomaly Detection with Contrastively-Learned Prompt (AAAI 2024 · similarity 0.85)
- 3CAD: A Large-Scale Real-World 3C Product Dataset for Unsupervised Anomaly Detection (AAAI 2025 · similarity 0.84)
- Towards Generalizable 3D Anomaly Detection via Relational Inconsistency Modeling (NeurIPS 2026 · similarity 0.83)
- FAST: Foreground‑aware Diffusion with Accelerated Sampling Trajectory for Segmentation‑oriented Anomaly Synthesis (NeurIPS 2025 · similarity 0.83)
- Multimodal Industrial Anomaly Detection by Crossmodal Feature Mapping (CVPR 2024 · similarity 0.83)
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.
