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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| L2-normalize features before the classifier (ASL) instead of standard CE | ResNet-18 · ImageNet-200 near-OOD | AUROC (%): 82.8 -> 85.1 | Improved |
| ASL instead of CE | ResNet-18 · CIFAR-100 near-OOD | AUROC (%): 74.8 -> 80.9 | Improved |
| ASL instead of CE | ResNet-18 · CIFAR-10 far-OOD | AUROC (%): 91.0 -> 97.2 | Improved |
| ASL instead of CE (fine-tuning) | ResNet-50 · ImageNet-1K near-OOD | AUROC (%): 71.4 -> 78.8 | Improved |
| ASL instead of CE (fine-tuning) | ViT-B/16 · ImageNet-1K far-OOD | AUROC (%): 91.9 -> 94.0 | Improved |
| ASL instead of CE | ResNet-50 · ImageNet-1K far-OOD | AUROC (%): 93.9 -> 93.3 | Hurt |
| ASL instead of CE under covariate shift | ResNet-18 · ImageNet-200 far-OOD | AUROC (%): 90.3 -> 90.8 | No clear effect |
| ASL instead of CE on medical data | ResNet-18 · BIMCV avg | FPR95 (%): 23.0 -> 12.4 | Improved |
| use ASL-trained features with MDS (ablation) | MDS · CIFAR-10 near-OOD | FPR95 (%): 49.2 -> 25.1 | Improved |
| use ASL-trained features with VIM (ablation) | VIM · CIFAR-10 near-OOD | FPR95 (%): 45.6 -> 26.2 | Improved |
| use ASL-trained features with cosine score (ablation) | FeatCos · CIFAR-10 near-OOD | FPR95 (%): 38.9 -> 30.5 | Improved |
| use ASL-trained features with cosine score (ablation) | FeatCos · CIFAR-100 near-OOD | AUROC (%): 80.3 -> 80.3 | No clear effect |
Closest papers by meaning
- Pursuing Feature Separation based on Neural Collapse for Out-of-Distribution Detection (ICLR 2025 · similarity 0.91)
- Detecting Out-of-Distribution Through the Lens of Neural Collapse (CVPR 2025 · similarity 0.91)
- MahaVar: OOD Detection via Class-wise Mahalanobis Distance Variance under Neural Collapse (NeurIPS 2026 · similarity 0.90)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · similarity 0.90)
- Is Complex Training Necessary for Long-Tailed OOD Detection? A Re-think from Feature Geometry (NeurIPS 2026 · similarity 0.90)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.90)
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.
