AI Paper Atlas

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

Why Feature Magnitude Deceives OOD Detectors: An Angular Separation Perspective

Hanlin Li, Jing Ma, Zehang Wei, Jiamin Yan, Xiang Xiang

ECCV 2026 · Label-efficient and robust learning › OOD and anomaly detection · approach: Provable and feature-based 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
L2-normalize features before the classifier (ASL) instead of standard CEResNet-18 · ImageNet-200 near-OODAUROC (%): 82.8 -> 85.1Improved
ASL instead of CEResNet-18 · CIFAR-100 near-OODAUROC (%): 74.8 -> 80.9Improved
ASL instead of CEResNet-18 · CIFAR-10 far-OODAUROC (%): 91.0 -> 97.2Improved
ASL instead of CE (fine-tuning)ResNet-50 · ImageNet-1K near-OODAUROC (%): 71.4 -> 78.8Improved
ASL instead of CE (fine-tuning)ViT-B/16 · ImageNet-1K far-OODAUROC (%): 91.9 -> 94.0Improved
ASL instead of CEResNet-50 · ImageNet-1K far-OODAUROC (%): 93.9 -> 93.3Hurt
ASL instead of CE under covariate shiftResNet-18 · ImageNet-200 far-OODAUROC (%): 90.3 -> 90.8No clear effect
ASL instead of CE on medical dataResNet-18 · BIMCV avgFPR95 (%): 23.0 -> 12.4Improved
use ASL-trained features with MDS (ablation)MDS · CIFAR-10 near-OODFPR95 (%): 49.2 -> 25.1Improved
use ASL-trained features with VIM (ablation)VIM · CIFAR-10 near-OODFPR95 (%): 45.6 -> 26.2Improved
use ASL-trained features with cosine score (ablation)FeatCos · CIFAR-10 near-OODFPR95 (%): 38.9 -> 30.5Improved
use ASL-trained features with cosine score (ablation)FeatCos · CIFAR-100 near-OODAUROC (%): 80.3 -> 80.3No clear effect

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