Research map / Label-efficient and robust learning / Domain adaptation and generalization
Intra- & Extra-Source Exemplar-Based Style Synthesis for Improved Domain Generalization
Yumeng Li, Dan Zhang, Margret Keuper, Anna Khoreva
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 random noise masking when training the encoder used for ISSA (ablation) | ISSA (HRNet) · CS to Dark Zurich | mIoU: 17.36 -> 26.10 | Improved |
| add random noise masking when training the encoder used for ISSA (ablation) | ISSA (HRNet) · CS to ACDC | mIoU: 44.63 -> 47.43 | Improved |
| add random noise masking when training the encoder used for ISSA (ablation) | ISSA (HRNet) · Cityscapes (source) | mIoU: 69.68 -> 69.48 | No clear effect |
| raise masking ratio from 25% to 50% (ablation) | masked noise encoder (patch 4) · Cityscapes reconstruction | FID: 1.41 -> 2.01 | Hurt |
| change mask patch size from 2 to 4 (ablation) | masked noise encoder (ratio 25%) · Cityscapes reconstruction | FID: 1.50 -> 1.41 | No clear effect |
| use noise map at 32x64 instead of 4x8 to 8x16 (ablation) | masked noise encoder · Cityscapes reconstruction | FID: 14.90 -> 2.30 | Improved |
| replace Cityscapes-trained encoder with BDD100K-trained encoder and generator (ablation) | ISSA (HRNet) · CS to ACDC average | mIoU: 50.1 -> 50.4 | No clear effect |
| add extra-source landscape exemplars (ESSA) to ISSA | ISSA (HRNet) · CS to Dark Zurich | mIoU: 27.24 -> 29.06 | Improved |
| add ISSA on top of RobustNet | RobustNet (DeepLab v3+) · CS to Dark Zurich | mIoU: 20.11 -> 23.09 | Improved |
| add ISSA on top of SHADE | SHADE (DeepLab v3+) · CS to Dark Zurich | mIoU: 25.37 -> 25.58 | Mixed |
| add ISSA on top of MixStyle | MixStyle (DeepLab v2) · CS to ACDC | mIoU: 36.97 -> 41.81 | Improved |
| add StyleMix augmentation instead of baseline | HRNet · CS to ACDC-Fog | mIoU: 58.68 -> 49.11 | Hurt |
Closest papers by meaning
- Adversarial Style Augmentation for Domain Generalized Urban-Scene Segmentation (NeurIPS 2022 · similarity 0.89)
- Exploring Semantic Consistency and Style Diversity for Domain Generalized Semantic Segmentation (AAAI 2025 · similarity 0.89)
- Learning Content-Enhanced Mask Transformer for Domain Generalized Urban-Scene Segmentation (AAAI 2024 · similarity 0.88)
- Learning Frequency-Adapted Vision Foundation Model for Domain Generalized Semantic Segmentation (NeurIPS 2024 · similarity 0.87)
- Style Blind Domain Generalized Semantic Segmentation via Covariance Alignment and Semantic Consistence Contrastive Learning (CVPR 2024 · similarity 0.87)
- START: A Generalized State Space Model with Saliency-Driven Token-Aware Transformation (NeurIPS 2024 · similarity 0.86)
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.
