Research map / Label-efficient and robust learning / OOD and anomaly detection
Understanding the Feature Norm for Out-of-Distribution Detection
Jaewoo Park, Jacky Chen Long Chai, Jaeho Yoon, Andrew Beng Jin Teoh
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 |
|---|---|---|---|
| add sparsity term to the l1 feature norm (NAN) (ablation) | feature norm on ResNet-50 with projection head · ImageNet-1k OOD (avg) | AUROC: 57.99 -> 92.32 | Improved |
| add sparsity term to the l1 feature norm (NAN) (ablation) | feature norm on ResNet-18 CIFAR-10 · CIFAR-10 OOD (avg) | AUROC: 92.40 -> 94.90 | Improved |
| add sparsity term when no projection head is used (ablation) | NAN on ResNet-50 average pool layer · ImageNet-1k OOD (avg) | AUROC: 87.09 -> 86.33 | Hurt |
| apply NAN on projection-head hidden layer instead of average pool (ablation) | NAN on ResNet-18 CIFAR-10 · CIFAR-10 OOD (avg) | FPR95: 38.86 -> 30.08 | Improved |
| combine NAN with KNN, SSD and ReAct | NAN · ImageNet-1k OOD (avg) | AUROC: 92.32 -> 94.61 | Improved |
| combine NAN with KNN in one-class classification | KNN · CIFAR-10 one-class | AUROC: 92.1 -> 94.3 | Improved |
| combine NAN with SSD in one-class classification | SSD · CIFAR-100 one-class | AUROC: 85.7 -> 88.7 | Improved |
| use NAN instead of distance-based detectors on far-OOD Texture | NAN · ImageNet-1k Texture | AUROC: 88.09 vs 96.96 (SSD) | Hurt |
| increase last hidden layer dimension (ablation) | NAN on ResNet-18 CIFAR-10 · CIFAR-10 OOD | AUROC/FPR95 | Improved |
| train with discriminative labels (supervised, instance, random) vs one label for all (ablation) | feature norm on ResNet-18 CIFAR-10 · CIFAR-10 OOD | AUROC | Improved |
Closest papers by meaning
- Detecting Out-of-Distribution Through the Lens of Neural Collapse (CVPR 2025 · similarity 0.88)
- Pursuing Feature Separation based on Neural Collapse for Out-of-Distribution Detection (ICLR 2025 · similarity 0.87)
- Mysteries of the Deep: Role of Intermediate Representations in Out of Distribution Detection (NeurIPS 2025 · similarity 0.87)
- CADRef: Robust Out-of-Distribution Detection via Class-Aware Decoupled Relative Feature Leveraging (CVPR 2025 · similarity 0.87)
- Exploiting Discrepancy in Feature Statistic for Out-of-Distribution Detection (AAAI 2024 · similarity 0.87)
- CASE: Exploiting Intra-class Compactness and Inter-class Separability of Feature Embeddings for Out-of-Distribution Detection (AAAI 2024 · similarity 0.87)
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.
