Research map / Label-efficient and robust learning / Domain adaptation and generalization
Split-and-Combine: Enhancing Style Augmentation for Single Domain Generalization
Zhen Zhang, Shuai Yang, Qianlong Dang, Zhize Wu, Lichuan Gu
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 ARC (adaptive random convolution) (ablation) | SAC · Digits | average accuracy: 77.7 -> 88.3 | Improved |
| add PAA (ablation) | SAC · Digits | average accuracy: 85.3 -> 88.3 | Improved |
| add Gaussian filter g (ablation) | SAC · Digits | average accuracy: 86.8 -> 88.3 | Improved |
| add feature-level discrepancy loss Ld (ablation) | SAC · Digits | average accuracy: 87.4 -> 88.3 | Improved |
| add pixel-level discrepancy loss Lp (ablation) | SAC · Digits | average accuracy: 87.0 -> 88.3 | Improved |
| use adaptive iteration strategy (Eq. 9) instead of random T in U[1,10] (ablation) | SAC · Digits | average accuracy: 86.9 -> 88.3 | Improved |
| use adaptive strategy instead of fixed T=10 (ablation) | SAC · Digits | average accuracy: 86.1 -> 88.3 | Improved |
| apply SAC to PDEN | PDEN · Digits | average accuracy: 74.8 -> 83.0 | Improved |
| apply SAC to L2D | L2D · Digits | average accuracy: 74.5 -> 76.4 | Improved |
| apply SAC to PSDG | PSDG · Digits | average accuracy: 84.7 -> 87.0 | Improved |
| apply ARC to Pro-RandConv | Pro-RandConv · Digits | average accuracy: 81.4 -> 83.6 | Improved |
| use deformable convolution instead of conventional convolution (ablation) | SAC · Digits | average accuracy: 86.4 -> 88.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AdvST — AdvST: Revisiting Data Augmentations for Single Domain Generalization (AAAI 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning To Diversify for Single Domain Generalization (ICCV 2021 · similarity 0.89)
- Style Neophile: Constantly Seeking Novel Styles for Domain Generalization (CVPR 2022 · similarity 0.88)
- StyleMix: Separating Content and Style for Enhanced Data Augmentation (CVPR 2021 · similarity 0.86)
- Style Evolving along Chain-of-Thought for Unknown-Domain Object Detection (CVPR 2025 · similarity 0.86)
- StyDeSty: Min-Max Stylization and Destylization for Single Domain Generalization (ICML 2024 · similarity 0.85)
- Style Adaptation and Uncertainty Estimation for Multi-Source Blended-Target Domain Adaptation (NeurIPS 2024 · similarity 0.85)
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.
