Research map / Label-efficient and robust learning / OOD and anomaly detection
Rethinking Out-of-Distribution Detection From a Human-Centric Perspective
Yao Tai Zhu, Yuefeng Chen, Xiaodan Li, Rong Zhang, Hui Xue, Xiang Tian, Rongxin Jiang, Bolun Zheng, Yaowu Chen
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 |
|---|---|---|---|
| evaluate with only correctly classified ID images instead of all validation images | MSP vs GradNorm on ResNet-50 · ImageNet OOD (iNaturalist, Places, SUN, Textures avg) | FPR95: 27.54% worse than GradNorm -> surpasses GradNorm by 16.55% | Improved |
| use MSP instead of ReAct | VGG19 · Human-centric OOD average | DER99: MSP surpasses ReAct by 1.87% | Improved |
| use MSP instead of ViM | VGG19 · Human-centric OOD average | DER99: MSP surpasses ViM by 36.5% | Improved |
| use ViM instead of MSP | Swin-B · Human-centric OOD average | DER99: MSP lags ViM by 0.14% | No clear effect |
| use MSP instead of GradNorm | Swin-B · Human-centric OOD average | DER99: MSP surpasses ReAct and GradNorm by 2.62% and 34.03% | Improved |
| use KNN instead of MSP | SENet154 vs ResNet-50 · Human-centric OOD average | DER99: KNN surpasses MSP by 4.9% on SENet154 but drops dramatically on ResNet50 | Mixed |
| use ViM on VGG19 versus Swin | ViM · iNaturalist | DER99: over 98% DER99 on VGG19 | Hurt |
| use larger ResNet-101 instead of ResNet-50 (ablation) | ViM · Human-centric OOD average | DER99: 6.67% better than ResNet50, 1.92% better than ResNet152 | Improved |
| increase adversarial training strength to epsilon=1 (ablation) | MSP on ResNet-50 · Human-centric OOD average | DER99: 15.32% worse than normal model | Hurt |
| use knowledge-distilled ResNet-50 (ablation) | KNN · Human-centric OOD average | DER99: improves by 26.95% | Improved |
| use knowledge-distilled DeiT-B (ablation) | KNN · Human-centric OOD average | DER99: degrades by 8.96% | Hurt |
| use knowledge-distilled models (ablation) | MSP · Human-centric OOD average | DER99 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ReAct — ReAct: Out-of-distribution Detection With Rectified Activations (NeurIPS 2021) · better on 5 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Understanding Failures in Out-of-Distribution Detection with Deep Generative Models (ICML 2021 · similarity 0.92)
- In or Out? Fixing ImageNet Out-of-Distribution Detection Evaluation (ICML 2023 · similarity 0.92)
- Unified Out-Of-Distribution Detection: A Model-Specific Perspective (ICCV 2023 · similarity 0.92)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.91)
- Image Background Serves as Good Proxy for Out-of-distribution Data (ICLR 2024 · similarity 0.91)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 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.
