Research map / Label-efficient and robust learning / OOD and anomaly detection
RankOOD - Class Ranking-based Out-of-Distribution Detection
Dishanika Denipitiyage, Naveen Karunanayake, Suranga Seneviratne, Sanjay Chawla
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 RankOOD-T (ListMLE ranking loss) and score with RankOOD-S instead of CE-trained MSP | RankOOD · CIFAR-10 (ID) vs SVHN (OOD) | FPR95 (%): 24.24 -> 12.23 (37.12% reduction) | Improved |
| RankOOD vs strongest baseline on near-OOD ImageNet-200 | RankOOD · Near-OOD, ImageNet-200 as ID | FPR95 (%): reduces FPR95 by 4.3% and improves AUROC by 0.50% | Improved |
| RankOOD vs prior rank-based methods CRAFT and ExCeL | RankOOD · Far-OOD and near-OOD averages | FPR95 (%): 7.51% far-OOD and 4.21% near-OOD average FPR95 reduction | Improved |
| RankOOD vs GEN among methods without outliers | RankOOD · Near-OOD, CIFAR-100 as ID | FPR95 (%): outperforms GEN by 3.36% | Improved |
| RankOOD vs G-ODIN on far-OOD TinyImageNet | RankOOD · Far-OOD, ImageNet-200 as ID | FPR95 (%): outperforms G-ODIN by 8.12% | Improved |
| RankOOD vs OE (trained with outliers) on near-OOD average | RankOOD · Near-OOD average | FPR95 (%): 44.79 vs 34.29 (OE better) | Hurt |
| train on top ranks only (vs adding bottom ranks) (ablation) | RankOOD-T · CIFAR-100 | ID accuracy | Hurt |
| add bottom ranks to top ranks (Top-Bottom, top 10 and bottom 10, N=20) (ablation) | RankOOD-T · CIFAR-100 | AUROC and ID accuracy | Improved |
| use more ranks in Top-Bottom subset beyond N=20 (ablation) | RankOOD-T · CIFAR-100 | AUROC | Hurt |
| sweep loss weight alpha in CE + alpha*ListMLE (ablation) | RankOOD-T · CIFAR-10 and CIFAR-100 | AUROC: best alpha 0.8 for CIFAR-10 and 1.0 for CIFAR-100 | Mixed |
| raise reference logit percentile threshold Ref past 0.95 (ablation) | RankOOD-S · CIFAR-10 and CIFAR-100 | AUROC: AUROC peaks around 0.95 then declines | Mixed |
| train with RankOOD-T instead of standard CE under various logit-based OOD scores (ablation) | RankOOD-T · OpenOOD benchmark | AUROC | Improved |
Closest papers by meaning
- CASE: Exploiting Intra-class Compactness and Inter-class Separability of Feature Embeddings for Out-of-Distribution Detection (AAAI 2024 · similarity 0.85)
- Learning to Shape In-distribution Feature Space for Out-of-distribution Detection (NeurIPS 2024 · similarity 0.84)
- OOD-Chameleon: Is Algorithm Selection for OOD Generalization Learnable? (ICML 2025 · similarity 0.84)
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · similarity 0.84)
- Plugin estimators for selective classification with out-of-distribution detection (ICLR 2024 · similarity 0.84)
- Out-of-Distribution Detection with Prototypical Outlier Proxy (AAAI 2025 · similarity 0.84)
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.
