Research map / Label-efficient and robust learning / OOD and anomaly detection
Watermarking for Out-of-distribution Detection
Read the paperCheck 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 |
|---|---|---|---|
| Add a learned watermark to test-time inputs | Softmax OOD scoring · CIFAR-10 OOD average | FPR95: 55.70 -> 42.86 | Helped |
| Add a learned watermark to test-time inputs | Free-energy OOD scoring · CIFAR-10 OOD average | AUROC: 90.56 -> 95.21 | Helped |
| Add a learned watermark to test-time inputs | Softmax OOD scoring · CIFAR-100 OOD average | FPR95: 82.97 -> 79.98 | Helped |
| Add a learned watermark to test-time inputs | Free-energy OOD scoring · CIFAR-100 OOD average | FPR95: 81.61 -> 77.07 | Helped |
| Add a learned watermark to test-time inputs | Softmax OOD scoring · ImageNet OOD average | FPR95: 54.93 -> 40.50 | Helped |
| Add a learned watermark to test-time inputs | Free-energy OOD scoring · ImageNet OOD average | FPR95: 52.73 -> 43.23 | Helped |
| Set Gaussian perturbation sigma1 to 0.40 rather than 0.00 (ablation) | Softmax watermarking · CIFAR-10 OOD average | FPR95: 48.71 -> 38.66 | Helped |
| Set Gaussian perturbation sigma1 to 2.00 rather than 0.40 (ablation) | Softmax watermarking · CIFAR-10 OOD average | FPR95: 38.66 -> 42.11 | Hurt |
| Use SAM with rho 1.00 rather than rho 0.00 (ablation) | Softmax watermarking · CIFAR-10 OOD average | FPR95: 43.04 -> 39.12 | Helped |
| Increase SAM rho from 1.00 to 5.00 (ablation) | Softmax watermarking · CIFAR-10 OOD average | FPR95: 39.12 -> 60.02 | Hurt |
| Use rotate-shifted ID data as OOD examples during watermark training (ablation) | Softmax watermarking · CIFAR-10 near OOD (CIFAR-100 as OOD) | FPR95: 90.10 -> 81.50 | Helped |
| Use rotate-shifted ID data as OOD examples during watermark training (ablation) | Free-energy watermarking · CIFAR-10 near OOD (CIFAR-100 as OOD) | FPR95: 52.25 -> 47.85 | Helped |
Closest papers by meaning
- Splitting & Integrating: Out-of-Distribution Detection via Adversarial Gradient Attribution (ICML 2025 · similarity 0.88)
- Image Background Serves as Good Proxy for Out-of-distribution Data (ICLR 2024 · similarity 0.87)
- Exploiting Discrepancy in Feature Statistic for Out-of-Distribution Detection (AAAI 2024 · similarity 0.86)
- What If the Input is Expanded in OOD Detection? (NeurIPS 2024 · similarity 0.86)
- Safe and Robust Watermark Injection with a Single OoD Image (ICLR 2024 · similarity 0.86)
- Learning by Erasing: Conditional Entropy Based Transferable Out-of-Distribution Detection (AAAI 2024 · similarity 0.86)
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.