Research map / Label-efficient and robust learning / OOD and anomaly detection
Diagnosing Pretrained Models for Out-of-distribution Detection
Haipeng Xiong, Kai Xu, Angela Yao
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 |
|---|---|---|---|
| train with the v2 recipe (label smoothing, mixup, etc.) instead of v1 | ResNet50 · ImageNet-1k | OOD AUROC: 4% ... 12% | Hurt |
| add mixup to the v1 recipe (ablation) | ResNet18 v1 · ImageNet200 | AUROC: 87.00 -> 84.00 | Hurt |
| replace mixup with regmixup (ablation) | ResNet18 v1 · ImageNet200 | AUROC: 87.00 -> 86.97 | No clear effect |
| use mixup-AugRevise with virtual separation loss (ablation) | ResNet18 v1 · ImageNet200 | AUROC: 87.00 -> 87.72 | Improved |
| apply AugDelete (refit last layer) to v2* (ablation) | ResNet18 v2* · ImageNet200 | AUROC: 82.78 -> 85.57 | Improved |
| train with AugRevise instead of v2* recipe | ResNet18 · ImageNet200 | AUROC: 82.78 -> 87.88 | Improved |
| add label smoothing to AugRevise (ablation) | AugRevise · ImageNet200 | AUROC: 87.88 -> 87.17 | Hurt |
| apply AugDelete with MLS score | ResNet50-v2 · ImageNet-1k | AUROC: 72.84 -> 83.08 | Improved |
| apply AugDelete with ASH score | ResNet50-v2 · ImageNet-1k | AUROC: 53.53 -> 81.70 | Improved |
| apply AugDelete with KNN score | ResNet50-v2 · ImageNet-1k | AUROC: 79.91 -> 79.91 | No clear effect |
| apply AugDelete with NNGuide score | ResNet50-v2 · ImageNet-1k | AUROC: 65.77 -> 77.54 | Improved |
| use AugRevise training with NNGuide score instead of v1 | ResNet18 · ImageNet200 | AUROC: 87.83 -> 89.31 | Improved |
Closest papers by meaning
- Revisiting Logit Distributions for Reliable Out-of-Distribution Detection (NeurIPS 2025 · similarity 0.91)
- Enhancing the Power of OOD Detection via Sample-Aware Model Selection (CVPR 2024 · similarity 0.91)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · similarity 0.90)
- Rethinking the Evaluation of Out-of-Distribution Detection: A Sorites Paradox (NeurIPS 2024 · similarity 0.90)
- Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection (NeurIPS 2024 · similarity 0.89)
- Can We Ignore Labels in Out of Distribution Detection? (ICLR 2025 · similarity 0.89)
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.
