Research map / Label-efficient and robust learning / Domain adaptation and generalization
Causal-Tune: Mining Causal Factors from Vision Foundation Models for Domain Generalized Semantic Segmentation
Yin Zhang, Yongqiang Zhang, Yaoyue Zheng, Bogdan Raducanu, Dan Liu
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 DCT instead of FFT for the frequency transform (ablation) | Causal-Tune · Cityscapes -> ACDC (avg) | mIoU: 69.5 -> 72.0 | Improved |
| use DCT instead of Haar wavelet transform (ablation) | Causal-Tune · Cityscapes -> ACDC (avg) | mIoU: 69.2 -> 72.0 | Improved |
| use DCT instead of FFT (ablation) | Causal-Tune · Cityscapes -> Mapillary | mIoU: 71.00 -> 76.05 | Improved |
| use DCT instead of FFT (ablation) | Causal-Tune · ACDC Night | mIoU: 57.1 -> 56.2 | Hurt |
| use band-pass filter instead of only removing low frequencies (ablation) | Causal-Tune · Cityscapes -> ACDC (avg) | mIoU: 68.4 -> 72.0 | Improved |
| use band-pass filter instead of only removing high frequencies (ablation) | Causal-Tune · Cityscapes -> ACDC (avg) | mIoU: 70.4 -> 72.0 | Improved |
| remove only high frequencies of DCT spectrum vs no filtering (ablation) | Rein baseline · Cityscapes -> ACDC (avg) | mIoU: 69.6 -> 70.4 | Improved |
| remove only low frequencies of DCT spectrum vs no filtering (ablation) | Rein baseline · Cityscapes -> ACDC (avg) | mIoU: 69.6 -> 68.4 | Hurt |
| add Causal-Tune to Rein baseline | Rein (DINOv2) · Cityscapes -> ACDC Snow | mIoU: 70.6 -> 75.4 | Improved |
| add Causal-Tune to Rein baseline | Rein (DINOv2) · GTA5 -> Cityscapes | mIoU: 66.40 -> 66.22 | No clear effect |
| add Causal-Tune to Rein baseline | Rein (DINOv2) · Cityscapes -> ACDC Night | mIoU: 55.9 -> 56.2 | No clear effect |
| raise both band-pass cutoff frequencies above RL 0.2 and RH 0.8 (ablation) | Causal-Tune · Cityscapes -> ACDC | mIoU | Hurt |
Closest papers by meaning
- FisherTune: Fisher-Guided Robust Tuning of Vision Foundation Models for Domain Generalized Segmentation (CVPR 2025 · similarity 0.90)
- Stronger Fewer & Superior: Harnessing Vision Foundation Models for Domain Generalized Semantic Segmentation (CVPR 2024 · similarity 0.89)
- Learning Frequency-Adapted Vision Foundation Model for Domain Generalized Semantic Segmentation (NeurIPS 2024 · similarity 0.87)
- Collaborating Foundation Models for Domain Generalized Semantic Segmentation (CVPR 2024 · similarity 0.87)
- Adapter Naturally Serves as Decoupler for Cross-Domain Few-Shot Semantic Segmentation (ICML 2025 · similarity 0.87)
- Leveraging Depth and Language for Open-Vocabulary Domain-Generalized Semantic Segmentation (NeurIPS 2025 · similarity 0.87)
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.
