Research map / Label-efficient and robust learning / Test-time adaptation
Test-Time Style Shifting: Handling Arbitrary Styles in Domain Generalization
Jungwuk Park, Dong-Jun Han, Soyeong Kim, Jaekyun Moon
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 test-time style shifting to MixStyle | MixStyle · PACS original | average accuracy: 82.98 -> 84.28 | Improved |
| Add style balancing and test-time style shifting to MixStyle | MixStyle · PACS cross-domain data imbalance | average accuracy: 75.65 -> 80.69 | Improved |
| Add style balancing to DSU | DSU · PACS cross-domain class imbalance | average accuracy: 40.28 -> 51.75 | Improved |
| Add test-time style shifting to DSU without style balancing | DSU · PACS cross-domain class imbalance | average accuracy: 40.28 -> 39.47 | Hurt |
| Add style balancing and test-time style shifting to DSU | DSU · PACS cross-domain class imbalance | average accuracy: 40.28 -> 52.22 | Improved |
| Add style balancing and test-time style shifting to MLDG | MLDG · PACS cross-domain class imbalance | average accuracy: 40.26 -> 53.54 | Improved |
| Add style balancing and test-time style shifting to CDANN | CDANN · PACS cross-domain class imbalance | average accuracy: 25.61 -> 46.22 | Improved |
| Select style-balancing samples by similar style statistics instead of randomly (ablation) | Style balancing + MixStyle · PACS cross-domain class imbalance | average accuracy: 49.51 -> 50.32 | Improved |
| Select style-balancing samples by similar style statistics instead of randomly (ablation) | TSB + MixStyle · PACS cross-domain class imbalance | average accuracy: 49.32 -> 50.83 | Improved |
| Shift all test samples, including those with small style gaps (ablation) | TSB + MixStyle · PACS original | average accuracy: 84.63 -> 84.68 | No clear effect |
| Shift to the nearest sample instead of the nearest source-domain center (ablation) | TSB + MixStyle · PACS original | average accuracy: 84.63 -> 84.14 | Hurt |
| Apply style shifting after the 4th residual block rather than the 2nd (ablation) | TSB + MixStyle · PACS original | average accuracy: 84.63 -> 20.21 | Hurt |
Closest papers by meaning
- Improved Test-Time Adaptation for Domain Generalization (CVPR 2023 · similarity 0.90)
- Learning To Diversify for Single Domain Generalization (ICCV 2021 · similarity 0.89)
- Improving Domain Generalization with Domain Relations (ICLR 2024 · similarity 0.87)
- Is Large-scale Pretraining the Secret to Good Domain Generalization? (ICLR 2025 · similarity 0.86)
- StyDeSty: Min-Max Stylization and Destylization for Single Domain Generalization (ICML 2024 · similarity 0.86)
- TTA-FedDG: Leveraging Test-Time Adaptation to Address Federated Domain Generalization (AAAI 2025 · 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.
