Research map / Label-efficient and robust learning / OOD and anomaly detection
Model-Free Test Time Adaptation for Out-of-Distribution Detection
Yi-Fan Zhang, Xue Wang, Tian Zhou, Kun Yuan, Zhang Zhang, Liang Wang, Rong Jin
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 test-time memory augmentation with detected OOD samples to KNN | KNN detector · CIFAR-10 (CE-trained) | FPR95 avg: 30.77 -> 7.24 | Improved |
| add test-time memory augmentation with detected OOD samples to KNN | KNN detector · ImageNet-1k | FPR95 avg: 53.97 -> 15.97 | Improved |
| use supervised contrastive training instead of cross-entropy | AdaODD · CIFAR-10 | FPR95 avg: 7.24 -> 5.51 | Improved |
| convert KNN into gradient-based TTA with triplet loss | KNN · CIFAR-10 (4 OOD sets) | FPR95 avg: 25.74 -> 20.76 | Improved |
| convert Mahalanobis into gradient-based TTA with triplet loss | Mahalanobis · CIFAR-10 (4 OOD sets) | FPR95 avg: 39.07 -> 33.84 | Improved |
| convert Energy into gradient-based TTA with hinge loss | Energy · CIFAR-10 (4 OOD sets) | FPR95 avg: 36.84 -> 38.91 | Hurt |
| raise selection margin gamma from 1.0 to 1.2 (fewer test samples added) (ablation) | AdaODD · ImageNet-1k | FPR95 avg: 15.97 -> 37.81 | Hurt |
| use gamma=infinity, adding no test samples (ablation) | AdaODD · ImageNet-1k | FPR95 avg: 15.97 -> 54.32 | Hurt |
| raise gamma from 1.0 to infinity increases ID reevaluation accuracy (ablation) | AdaODD · ImageNet-1k | ID reevaluation accuracy avg: 93.10 -> 95.00 | Mixed |
| add feature normalization before kNN search (ablation) | AdaODD · ImageNet-1k | FPR avg: +66.2 | Improved |
| feed OOD datasets sequentially (Stream) | AdaODD · CIFAR-10 Texture | FPR95: 15.82 -> 3.83 | Improved |
| mix several OOD datasets into one test stream | AdaODD · CIFAR-10 SVHN | FPR95: 2.17 -> 1.13 | 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.
- Mahalanobis — Mahalanobis++: Improving OOD Detection via Feature Normalization (ICML 2025) · better on 1 of 1 shared result
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
- Enhancing the Power of OOD Detection via Sample-Aware Model Selection (CVPR 2024 · similarity 0.92)
- Test-Time Linear Out-of-Distribution Detection (CVPR 2024 · similarity 0.92)
- OODD: Test-time Out-of-Distribution Detection with Dynamic Dictionary (CVPR 2025 · similarity 0.90)
- Long-Tailed Out-of-Distribution Detection via Normalized Outlier Distribution Adaptation (NeurIPS 2024 · similarity 0.89)
- Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection (NeurIPS 2024 · similarity 0.89)
- Rethinking the Evaluation of Out-of-Distribution Detection: A Sorites Paradox (NeurIPS 2024 · 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.
