AI Paper Atlas

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

Is Task-Specific Training Necessary for Anomaly Detection?

Xingwu Zhang, Guanxuan Li, Paul Henderson, Gerardo Aragon-Camarasa, ZIJUN LONG

ICML 2026 · Label-efficient and robust learning › OOD and anomaly detection · approach: Unified multi-class anomaly detection

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
add global [CLS] retrieval before patch matching (ablation)RAD · MVTec-ADinference latency: 293.7 ms -> 44.7 msImproved
compress memory bank with coreset sampling (ablation)RAD · MVTec-ADpixel AP: 75.60 -> 73.86Hurt
compress memory bank with coreset sampling (ablation)RAD · MVTec-ADpixel AUPRO: 94.90 -> 94.75No clear effect
increase anomaly-free data ratio from 0.05 to 1.00 (single-class)RAD · MVTec-ADimage AUROC: 97.1 -> 99.6Improved
increase anomaly-free data ratio from 0.05 to 1.00 (single-class)RAD · MVTec-ADpixel AP: 70.6 -> 75.7Improved
use retrieval instead of trained decoder at tau=0.05 (single-class)RAD vs Dinomaly · MVTec-ADpixel AP: 54.7 -> 70.6Improved
use retrieval instead of trained decoder at tau=0.05 (multi-class)RAD vs Dinomaly · MVTec-ADpixel AP: 51.1 -> 67.9Improved
use retrieval instead of trained decoder with 2 transistor imagesRAD vs Dinomaly · MVTec-AD incrementalpixel AP: 33.1 -> 53.7Improved
use retrieval instead of trained decoderRAD vs Dinomaly · MVTec-AD few-shotpixel AUPRO: +16.8 / +16.9 / +14.9 at Y=1,2,4Improved
use RAD instead of few-shot method IIPADRAD vs IIPAD · MVTec-AD few-shotpixel AUPRO: +2.1 / +2.7 / +2.7 at Y=1,2,4Improved
reduce full data to a single anomaly-free image (ablation)RAD · MVTec-ADpixel AUROC: 98.5 -> 96.7Hurt
swap in stronger frozen encoder (DINOv3 vs MIM encoders) (ablation)RAD · MVTec-ADP-AUPROImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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