Research map / Label-efficient and robust learning / OOD and anomaly detection
Key Feature Replacement of In-Distribution Samples for Out-of-Distribution Detection
Jaeyoung Kim, Seo Taek Kong, Dongbin Na, Kyu-Hwan Jung
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 |
|---|---|---|---|
| Use KIRBY-B instead of the strongest listed baseline | CIFAR-10 OOD detection · Six OOD benchmarks | AUROC: 95.33 -> 97.82 | Improved |
| Use KIRBY-B rather than OE | CIFAR-10 OOD detection · Six OOD benchmarks | AUROC: 98.34 -> 97.82 | Hurt |
| Use KIRBY-B rather than OE | CIFAR-100 OOD detection · Six OOD benchmarks | AUROC: 84.60 -> 93.11 | Improved |
| Use KIRBY-B rather than the strongest listed baseline | CIFAR-100 OOD detection · Six OOD benchmarks | AUROC: 86.09 -> 93.11 | Improved |
| Use KIRBY-M rather than the strongest listed baseline | STL-10 OOD detection · Six OOD benchmarks | AUROC: 84.83 -> 91.25 | Improved |
| Add CAM-guided key-feature erasure instead of randomly masking image regions (ablation) | KIRBY-M · CIFAR-10 | AUROC: 95.16 -> 95.30 | Improved |
| Add inpainting after erasure (ablation) | KIRBY-M · CIFAR-10 | AUROC: 95.30 -> 96.33 | Improved |
| Add crop augmentation to the inpainted surrogate set (ablation) | KIRBY-M · CIFAR-10 | AUROC: 96.33 -> 97.03 | Improved |
| Add inpainting and crop augmentation to erasure (ablation) | KIRBY-M · CIFAR-100 | AUROC: 88.10 -> 92.31 | Improved |
| Add inpainting and crop augmentation to erasure (ablation) | KIRBY-M · STL-10 | AUROC: 83.35 -> 89.50 | Improved |
| Replace no inpainting with mean-value inpainting (ablation) | KIRBY-M · CIFAR-10 | AUROC: 95.30 -> 96.59 | Improved |
| Replace fast marching inpainting with conditional sampling (ablation) | KIRBY-M · STL-10 | AUROC: 89.50 -> 84.59 | Hurt |
Closest papers by meaning
- Image-based Outlier Synthesis With Training Data (CVPR 2026 · similarity 0.92)
- Image Background Serves as Good Proxy for Out-of-distribution Data (ICLR 2024 · similarity 0.91)
- ID-like Prompt Learning for Few-Shot Out-of-Distribution Detection (CVPR 2024 · similarity 0.91)
- Rethinking the Evaluation of Out-of-Distribution Detection: A Sorites Paradox (NeurIPS 2024 · similarity 0.91)
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.90)
- Exploiting Discrepancy in Feature Statistic for Out-of-Distribution Detection (AAAI 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.
