AI Paper Atlas

Research map / Label-efficient and robust learning / OOD and anomaly detection

Anomaly Detection using Score-based Perturbation Resilience

Woosang Shin, Jonghyeon Lee, Taehan Lee, Sangmoon Lee, Jong Pil Yun

ICCV 2023 · Label-efficient and robust learning › OOD and anomaly detection · approach: Unified multi-class anomaly detection · 27 citations (OpenAlex)

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 changedWhereMeasuredEffect
use score-based perturbation resilience without pre-trained models instead of the best prior unsupervised method NSAOurs · MVTec AD localizationpixel-level AUROC (%): 96.3 -> 97.4Improved
use score-based method instead of RIAD reconstruction methodOurs · MVTec AD localizationpixel-level AUROC (%): 94.2 -> 97.4Improved
use external knowledge (pre-trained model) like PatchCoreanomaly localization on MVTec AD · MVTec AD localizationpixel-level AUROC (%): 97.4 -> 98.1Improved
use score-based method instead of NSA on image-level detectionOurs · MVTec AD detectionimage-level AUROC (%): 97.2 -> 97.7Improved
increase number of predictions N from 1 to 3 (ablation)Ours · MVTec ADimage-level AUROC (%): 97.5 -> 97.7Improved
increase number of predictions N from 3 to 5 (ablation)Ours · MVTec ADimage-level AUROC (%): 97.7 -> 97.7No clear effect
tune the variance scale k per category rather than fixed k = 0.001 (ablation)Ours · MVTec AD detectionimage-level AUROC (%): 97.7 -> 98.9Improved
use too small or too large perturbation variance scale (ablation)Ours · perturbed data distributionqualitativeHurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy