Research map / Label-efficient and robust learning / OOD and anomaly detection
Full-Spectrum Out-of-Distribution Detection
Jingkang Yang, Kaiyang Zhou, Ziwei Liu
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 low-layer feature-statistics term p(x_n) to p(x) (SEM vs p(x) only) (ablation) | SEM · FS-OOD DIGITS far-OOD | AUROC: 60.80 -> 69.73 | Improved |
| add low-layer feature-statistics term p(x_n) to p(x) (SEM vs p(x) only) (ablation) | SEM · FS-OOD DIGITS near-OOD | AUROC: 87.28 -> 88.47 | Improved |
| use low-layer feature statistics instead of low-layer flattened features for p(x_n) (ablation) | SEM · FS-OOD DIGITS near-OOD | AUROC: 86.54 -> 88.47 | Improved |
| use low-layer feature statistics instead of low-layer flattened features for p(x_n) (ablation) | SEM · FS-OOD DIGITS far-OOD | AUROC: 61.26 -> 69.73 | Improved |
| add fine-tuning with L_cls before fitting the GMM (ablation) | SEM · FS-OOD DIGITS near-OOD | AUROC: 83.03 -> 86.55 | Improved |
| add L_src(x) (compact ID statistics) to fine-tuning (ablation) | SEM · FS-OOD DIGITS far-OOD | AUROC: 64.61 -> 68.40 | Improved |
| add negative augmented data L_src(x') to fine-tuning with L_src(x) (ablation) | SEM · FS-OOD DIGITS near-OOD | AUROC: 87.42 -> 88.47 | Improved |
| add L_src(x) to fine-tuning that already has negative augmentation (ablation) | SEM · FS-OOD DIGITS far-OOD | AUROC: 67.92 -> 69.73 | Improved |
| use M=3 GMM components for p(x_n) and M=10 for p(x) instead of M=1/M=20 (ablation) | SEM · FS-OOD DIGITS near-OOD | AUROC: 84.47 -> 88.47 | Improved |
| SEM vs best baseline on notMNIST near-OOD (feature-based semantic score) | SEM · FS-OOD DIGITS notMNIST | FPR95: 99.99 -> 10.93 | Improved |
| use SEM instead of EBO/MSP/ODIN under covariate-shifted ID | SEM · FS-OOD COVID near-OOD | AUROC: 45.47 -> 97.24 | Improved |
| use SEM instead of output-based scores on far-OOD of OBJECTS | SEM · FS-OOD OBJECTS far-OOD | FPR95: 76.98 -> 89.87 | Hurt |
Closest papers by meaning
- ImageNet-OOD: Deciphering Modern Out-of-Distribution Detection Algorithms (ICLR 2024 · similarity 0.94)
- Rethinking the Evaluation of Out-of-Distribution Detection: A Sorites Paradox (NeurIPS 2024 · similarity 0.94)
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · similarity 0.90)
- MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities (NeurIPS 2024 · similarity 0.89)
- Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection (NeurIPS 2024 · similarity 0.89)
- Image Background Serves as Good Proxy for Out-of-distribution Data (ICLR 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.
