Research map / Label-efficient and robust learning / Domain adaptation and generalization
Customizing Domain Adapters for Domain Generalization
Yuyang Ji, Zeyi Huang, Haohan Wang, Yong Jae Lee
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 router loss (ablation) | 3 ViT adapters · PACS | accuracy (%): 88.0 -> 88.5 | Improved |
| add router loss (ablation) | 3 Conv adapters · PACS | accuracy (%): 88.3 -> 88.7 | Improved |
| add router loss (ablation) | 3 customized adapters · PACS | accuracy (%): 88.9 -> 89.4 | Improved |
| add router loss (ablation) | 5 customized adapters · DomainNet | accuracy (%): 49.6 -> 50.3 | Improved |
| combine ViT and Conv adapters (customized) instead of one adapter type (ablation) | 2 adapters · PACS | accuracy (%): 87.9 -> 88.9 | Improved |
| increase number of identical ViT adapters from 2 to 3 (ablation) | ViT adapter · PACS | accuracy (%): 87.9 -> 88.0 | No clear effect |
| increase number of customized adapters from 2 to 3 without router loss (ablation) | customized adapter · PACS | accuracy (%): 88.9 -> 88.9 | No clear effect |
| choose adapters with GPT-4 instead of empirical selection (ablation) | CDA · PACS | accuracy (%): 89.4 -> 89.3 | No clear effect |
| add CDA architecture on top of ViT-S ERM | ViT-S/16 ERM · PACS | accuracy (%): 86.2 -> 89.4 | Improved |
| add CDA architecture on top of ViT-B ERM | ViT-B/16 ERM · PACS | accuracy (%): 88.8 -> 90.1 | Improved |
| add CDA to SWAD | SWAD · PACS | accuracy (%): 88.1 -> 90.3 | Improved |
| add CDA to W2D | W2D · PACS | accuracy (%): 86.6 -> 89.9 | Improved |
Closest papers by meaning
- Learning to Optimize Domain Specific Normalization for Domain Generalization (ECCV 2020 · similarity 0.90)
- Adaptive Methods for Real-World Domain Generalization (CVPR 2021 · similarity 0.89)
- Feature-Critic Networks for Heterogeneous Domain Generalization (ICML 2019 · similarity 0.89)
- Learning to Generate Novel Domains for Domain Generalization (ECCV 2020 · similarity 0.89)
- Learning To Diversify for Single Domain Generalization (ICCV 2021 · similarity 0.89)
- Exploiting Domain-Specific Features to Enhance Domain Generalization (NeurIPS 2021 · similarity 0.88)
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.
