Research map / Label-efficient and robust learning / OOD and anomaly detection
MOS: Towards Scaling Out-of-Distribution Detection for Large Semantic Space
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use MOS instead of KL Matching for OOD scoring | BiT-S-R101x1 ImageNet-1k classifier · Four OOD datasets, average | FPR95: 54.30 -> 39.97 | Helped |
| Use MOS instead of KL Matching for OOD scoring | BiT-S-R101x1 ImageNet-1k classifier · 60k in- and out-of-distribution images | test time (min): 20.6 -> 3.2 | Helped |
| Use feature clustering instead of taxonomy grouping (ablation) | MOS · Four OOD datasets | FPR95: outperforms MSP by 18.2% (FPR95) on average | Mixed |
| Use random grouping instead of taxonomy grouping (ablation) | MOS · Four OOD datasets | FPR95 | Mixed |
| Increase the number of ImageNet in-distribution classes from 50 to 1,000 (ablation) | MOS versus MSP · iNaturalist | FPR95: MSP rises from 21.02% to 63.36%; MOS degrades by only 4.76% | Helped |
| Increase the number of taxonomy groups from 8 to 36 (ablation) | MOS · Four OOD datasets, average | FPR95: 39.97 -> 42.33 | Hurt |
| Increase the number of taxonomy groups from 8 to 789 (ablation) | MOS · Four OOD datasets, average | FPR95: 39.97 -> 81.42 | Hurt |
| Fine-tune more residual blocks beyond the final fully connected layer (ablation) | MOS · OOD detection, ImageNet-1k | FPR95 | Hurt |
| Use higher-capacity pre-trained feature extractors (ablation) | MOS · OOD detection, ImageNet-1k | OOD detection performance | Helped |
Closest papers by meaning
- A Systematic Analysis of Out-of-Distribution Detection Under Representation and Training Paradigm Shifts (NeurIPS 2026 · similarity 0.91)
- ImageNet-OOD: Deciphering Modern Out-of-Distribution Detection Algorithms (ICLR 2024 · similarity 0.91)
- Image Background Serves as Good Proxy for Out-of-distribution Data (ICLR 2024 · similarity 0.91)
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- Rethinking the Evaluation of Out-of-Distribution Detection: A Sorites Paradox (NeurIPS 2024 · similarity 0.90)
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · 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.