Research map / Label-efficient and robust learning / Domain adaptation and generalization
Style-Hallucinated Dual Consistency Learning: A Unified Framework for Visual Domain Generalization
Yuyang Zhao, Zhun Zhong, Na Zhao, Nicu Sebe, Gim Hee Lee
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 style hallucination module (SHM) with plain CE on hallucinated samples (ablation) | ResNet-50 baseline · GTAV to real avg (ResNet-50 segmentation) | mIoU: 27.42 -> 35.68 | Improved |
| add style consistency (SC) on top of SHM (ablation) | SHM · GTAV to real avg (ResNet-50 segmentation) | mIoU: 35.68 -> 39.67 | Improved |
| add retrospection consistency (RC) on top of SHM (ablation) | SHM · GTAV to real avg (ResNet-50 segmentation) | mIoU: 35.68 -> 40.29 | Improved |
| add RC (ImageNet pre-trained model) to SHM+SC (ablation) | SHADE · GTAV to real avg (ResNet-50 segmentation) | mIoU: 39.67 -> 42.42 | Improved |
| use EMA model instead of ImageNet pre-trained model for RC (ablation) | SHM+SC · GTAV to real avg (ResNet-50 segmentation) | mIoU: 42.42 -> 40.92 | Hurt |
| use FPS basis styles instead of Kmeans basis styles in SHM (ablation) | SHADE · GTAV to real avg (style variation) | mIoU: 39.19 -> 42.42 | Improved |
| use SHM instead of MixStyle for style generation (ablation) | SHADE · GTAV to real avg (style variation) | mIoU: 41.10 -> 42.42 | Improved |
| use Kmeans basis instead of directly swapping/mixing source styles (ablation) | SHADE · GTAV to real avg (style variation) | mIoU: 41.10 -> 39.19 | Hurt |
| insert SHM after layer3 instead of layer0 (ablation) | SHADE · GTAV to real (ResNet-50 segmentation) | mIoU | Hurt |
| select basis styles only once instead of every k<=3 epochs (ablation) | SHADE · GTAV to real (ResNet-50 segmentation) | mIoU: 42.42 -> lower than 41 | Hurt |
| add SHM to ERM on PACS (ablation) | ERM · PACS avg | accuracy: 58.9 -> 59.6 | Improved |
| add style consistency on top of SHM on PACS (ablation) | SHM · PACS avg | accuracy: 59.6 -> 64.2 | 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.
- FSDR — FSDR: Frequency Space Domain Randomization for Domain Generalization (CVPR 2021) · 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
- Style-Hallucinated Dual Consistency Learning for Domain Generalized Semantic Segmentation (ECCV 2022 · similarity 0.94)
- Exploring Semantic Consistency and Style Diversity for Domain Generalized Semantic Segmentation (AAAI 2025 · similarity 0.88)
- Style Blind Domain Generalized Semantic Segmentation via Covariance Alignment and Semantic Consistence Contrastive Learning (CVPR 2024 · similarity 0.87)
- Learning Fine-grained Domain Generalization via Hyperbolic State Space Hallucination (AAAI 2025 · similarity 0.86)
- OmniConsistency: Learning Style-Agnostic Consistency from Paired Stylization Data (NeurIPS 2025 · similarity 0.86)
- DGFamba: Learning Flow Factorized State Space for Visual Domain Generalization (AAAI 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.
