Research map / Label-efficient and robust learning / Domain adaptation and generalization
ASTAD: Asymmetric Style Transfer for Synthetic-to-Real Adaptation in Autonomous Driving
Dingyi Yao, Xinqi Zhang, Lihui Peng, Jianming HU, Danya Yao, Yi ZHANG
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 semantically constrained adaptive attention filtering (ablation) | ASTModel · structural fidelity | LPIPS: 0.5114 -> 0.3588 | Improved |
| add pixel-proportion modulated hybrid AdaIN (vs global AdaIN) (ablation) | ASTModel · structural fidelity | LPIPS: 0.3737 -> 0.3588 | Improved |
| add multi-layer semantic voting (ablation) | ASTModel · structural fidelity | LPIPS: 0.3641 -> 0.3588 | Improved |
| refine style prior with multi-layer voting (Stage-II) (ablation) | ASTModel pseudo labels · style-side pseudo labels | Pixel Acc: 0.688 -> 0.790 | Improved |
| refine style prior with multi-layer voting (Stage-II) (ablation) | ASTModel pseudo labels · style-side pseudo labels | mIoU: 0.281 -> 0.339 | Improved |
| use Median/MAD instead of Mean/Std for filtering threshold (ablation) | ASTModel attention filtering · attention entries retained | fraction kept (%): 1.8% -> 33.2% | Improved |
| use ASTModel instead of CACTIF for synthetic-to-real transfer | downstream SegFormer · GTA to real downstream segmentation | mIoU: 0.289 -> 0.309 | Improved |
| use ASTModel instead of source-only training | downstream SegFormer · GTA to real downstream segmentation | Pixel Acc: 0.844 -> 0.847 | No clear effect |
| use ASTModel instead of CACTIF | style transfer · inference time per image | seconds: 80s -> 25s | Improved |
| use ASTModel instead of Cross-Image Attention | style transfer · inference time per image | seconds: 18s -> 25s | Hurt |
Closest papers by meaning
- Unpaired Geometry-Guided Sim2Real Translation for Autonomous Driving (ECCV 2026 · similarity 0.89)
- Towards Optimal Strategies for Training Self-Driving Perception Models in Simulation (NeurIPS 2021 · similarity 0.88)
- Rethinking Driving World Model as Synthetic Data Generator for Perception Tasks (ICLR 2026 · similarity 0.88)
- ReSimAD: Zero-Shot 3D Domain Transfer for Autonomous Driving with Source Reconstruction and Target Simulation (ICLR 2024 · similarity 0.86)
- Uncertainty-Guided Never-Ending Learning to Drive (CVPR 2024 · similarity 0.85)
- Transfer Your Perspective: Controllable 3D Generation from Any Viewpoint in a Driving Scene (CVPR 2025 · 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.
