Research map / Label-efficient and robust learning / Domain adaptation and generalization
Causal Inference via Style Bias Deconfounding for Domain Generalization
Jiaxi Li, Di Lin, Hao Chen, Hongying Liu, Liang Wan, Wei Feng
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 SDCL (style deconfounding causal learning) to SHADE | SHADE · Digits single-domain (train MNIST) | accuracy (%): 81.20 -> 82.07 | Improved |
| add SDCL (style deconfounding causal learning) to AdvST | AdvST · PACS single-domain | avg accuracy (%): 68.07 -> 71.00 | Improved |
| add SDCL (style deconfounding causal learning) to SHADE | SHADE · PACS single-domain | avg accuracy (%): 68.47 -> 70.40 | Improved |
| add SDCL (style deconfounding causal learning) to SHADE | SHADE · PACS multi-domain (leave-one-out) | avg accuracy (%): 85.31 -> 86.95 | Improved |
| add SDCL (style deconfounding causal learning) to TLDR | TLDR · GTAV to real (single-domain) | avg mIoU (%): 47.00 -> 48.47 | Improved |
| add SDCL (style deconfounding causal learning) to TLDR | TLDR · GTAV+SYNTHIA to real (multi-domain) | avg mIoU (%): 42.38 -> 44.40 | Improved |
| add SDCL (style deconfounding causal learning) to SHADE | SHADE · GTAV+SYNTHIA to real (multi-domain) | avg mIoU (%): 45.11 -> 46.90 | Improved |
| add SDCL (style deconfounding causal learning) to SLAug | SLAug · CDCS cardiac multi-domain (leave-one-out) | avg Dice (%): 84.80 -> 86.45 | Improved |
| add BDCL (back-door causal learning) on top of SGEM (ablation) | SDCL · PACS single-domain | avg accuracy (%): 69.83 -> 71.00 | Improved |
| add BDCL on top of SGEM (ablation) | SDCL · GTAV to real (single-domain) | avg mIoU (%): 47.93 -> 48.47 | Improved |
| add SGEM (style-guided expert module) to the base augmentation method (ablation) | UDP/AdvST/TLDR/SLAug · PACS single-domain | avg accuracy (%): 68.07 -> 69.83 | Improved |
| place SGEM at ResNet layer 1 instead of layer 4 (ablation) | SDCL · PACS single-domain | accuracy (%): 67.12 -> 71.00 | 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
- Test-Time Style Shifting: Handling Arbitrary Styles in Domain Generalization (ICML 2023 · similarity 0.88)
- Meta-Causal Learning for Single Domain Generalization (CVPR 2023 · similarity 0.87)
- Causality Inspired Representation Learning for Domain Generalization (CVPR 2022 · similarity 0.87)
- Explicit Modeling of Causal Factors and Confounders for Image Classification (AAAI 2026 · similarity 0.87)
- Causal-Tune: Mining Causal Factors from Vision Foundation Models for Domain Generalized Semantic Segmentation (AAAI 2026 · similarity 0.87)
- Towards domain-invariant Self-Supervised Learning with Batch Styles Standardization (ICLR 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.
