Research map / Label-efficient and robust learning / Domain adaptation and generalization
ConstStyle: Robust Domain Generalization with Unified Style Transformation
Nam Duong Tran, Nam Nguyen Phuong, Hieu H. Pham, Phi Le Nguyen, My T. Thai
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 pretrained features instead of learned style statistics for clustering (ablation) | ConstStyle · Digits5 | accuracy: 86.88 -> 76.61 | Hurt |
| use pretrained features instead of learned style statistics for clustering (ablation) | ConstStyle · PACS | accuracy: 86.77 -> 86.31 | No clear effect |
| use GMM clustering instead of domain labels (ablation) | ConstStyle · Digits5 | accuracy: 86.37 -> 86.88 | Improved |
| use GMM clustering instead of domain labels (ablation) | ConstStyle · PACS | accuracy: 86.73 -> 86.77 | No clear effect |
| use partial alignment alpha=0.6 instead of alpha=1.0 (ablation) | ConstStyle · PACS | accuracy: 85.86 -> 86.77 | Improved |
| use partial alignment alpha=0.5 instead of alpha=0 (full mapping) (ablation) | ConstStyle · Digits5 | accuracy: 85.96 -> 86.88 | Improved |
| use 4 clusters instead of 1 (ablation) | ConstStyle · Digits5 | accuracy: 85,80 -> 86.88 | Improved |
| use batch size 32 instead of 256 (ablation) | ConstStyle · PACS | accuracy: 85.10 -> 86.77 | Improved |
| use batch size 32 instead of 8 (ablation) | ConstStyle · PACS | accuracy: 85.43 -> 86.77 | Improved |
| apply ConstStyle unified-domain alignment vs ERM | ResNet18 · PACS (single unseen domain) | avg accuracy: 79.86 -> 86.77 | Improved |
| apply ConstStyle vs ERM at corruption level 5 | WideResNet · CIFAR10-C level 5 | accuracy: 47.80 -> 63.63 | Improved |
| apply ConstStyle vs ERM at corruption level 1 | WideResNet · CIFAR10-C level 1 | accuracy: 87.33 -> 86.59 | Hurt |
Closest papers by meaning
- DomainDrop: Suppressing Domain-Sensitive Channels for Domain Generalization (ICCV 2023 · similarity 0.89)
- Test-Time Style Shifting: Handling Arbitrary Styles in Domain Generalization (ICML 2023 · similarity 0.88)
- Seeking Consistent Flat Minima for Better Domain Generalization via Refining Loss Landscapes (CVPR 2025 · similarity 0.87)
- Set Valued Predictions For Robust Domain Generalization (ICML 2025 · similarity 0.87)
- StyDeSty: Min-Max Stylization and Destylization for Single Domain Generalization (ICML 2024 · similarity 0.87)
- Robust Domain Generalization under Divergent Marginal and Conditional Distributions (NeurIPS 2026 · 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.
