AI Paper Atlas

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

A Comprehensive Analysis about Unsupervised Outlier Detection for Images

Zhonghang Liu, Siyuan Chen, Jingwen Yu, Changshuo Wang, Kunyang Li, Jiangbo Lu

ECCV 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 VUOD feature enhancement (pseudo labels + contrastive learning) before OCSVM (ablation)OCSVM · CIFAR-10AUC: 0.824 -> 0.972Improved
add VUOD feature enhancement before OCSVM (ablation)OCSVM · MPDDAUC: 0.575 -> 0.916Improved
add VUOD feature enhancement before OCSVM (ablation)OCSVM · MVTec-ADAUC: 0.754 -> 0.921Improved
add VUOD feature enhancement before OCSVM (ablation)OCSVM · Caltech-101AUC: 0.955 -> 0.994Improved
add VUOD feature enhancement before OCSVM (ablation)OCSVM · BrainMRIAUC: 0.605 -> 0.797Improved
use Affinity Propagation instead of k-Means with the same cluster count (ablation)clustering stage · 10,000 ResNet-50 featuresseconds: 207.3 -> 84.60Improved
use refined distance (Eq. 6) instead of mean distance baseline (ablation)global separability · STL-10 across outlier ratiosAUCImproved
shift thresholds (eps+, eps-) from default (0.9, 0.5) (ablation)VUOD · CIFAR-10AUC: 0.972 -> 0.963No clear effect
filter training set with VUODResNet-18 · Imagenettetop-1 accuracy (%): 90.81% -> 91.45%Improved
filter training set with VUODResNet-18 · Imagenettetop-5 accuracy (%): 99.11% -> 99.39%Improved
use VUOD instead of RobustVGGT for outlier image filteringVGGT · 3D reconstruction input filteringms per image: 118 ms -> 27 msImproved
use VUOD instead of EfficientAD for defect scoringindustrial inspection · MVTec-AD bottleseconds: 42.10 -> 17.18Improved

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