AI Paper Atlas

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

Distribution-Aware Feature Selection for Post-hoc Out-of-Distribution Detection

Max Gutbrod, David Rauber, Christoph Palm

ECCV 2026 · Label-efficient and robust learning › OOD and anomaly detection · approach: Spurious-aware OOD detection · 0 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
add marginal feature selection (MFS)kNN · Average over 5 benchmarksAUROC: 75.83 -> 77.68Improved
add MFSMDS · Average over 5 benchmarksAUROC: 75.77 -> 78.20Improved
add MFSMDS++ · Average over 5 benchmarksAUROC: 79.25 -> 80.41Improved
add MFSkNN · MIDOGAUROC: 70.64 -> 76.66Improved
add MFSkNN · PhaKIRAUROC: 42.41 -> 42.41No clear effect
add MFSkNN · Average runtimeruntime reduction: 58.89 %Improved
add whitening before MFSkNN+MFS · PhaKIRAUROC: about 14 pointsImproved
use Wasserstein-1 instead of JS divergence as shift measure (ablation)kNN+MFS · Medical avgAUROC: 71.96 -> 72.87Improved
use mixup proxy-OOD instead of adversarial (ablation)MDS+MFS · Medical avgAUROC: 76.42 -> 79.14Improved
use mixup proxy-OOD instead of adversarial (ablation)MDS+MFS · Natural avgAUROC: 77.19 -> 76.79Hurt
use random feature subsets instead of MFS (ablation)MDS+MFS · MIDOG and ImageNet-1kAUROCImproved
use real-OOD validation for selection (ablation)kNN+MFS · Average over 5 benchmarksAUROC: +2.41Improved

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