Research map / Label-efficient and robust learning / Domain adaptation and generalization
Learning Texture Invariant Representation for Domain Adaptation of Semantic Segmentation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 texture-diversified Style-swap source training instead of training on original source images only (ablation) | Stage 1 segmentation training · GTA5 → Cityscapes | mIoU: 36.6 -> 42.5 | Helped |
| Add Style-swap stylization instead of using CycleGAN-translated source images only (ablation) | Stage 1 segmentation training · GTA5 → Cityscapes | mIoU: 41.0 -> 42.5 | Helped |
| Add adversarial loss to stylized-source training (ablation) | Stage 1 segmentation training · GTA5 → Cityscapes | mIoU: 42.5 -> 43.2 | Helped |
| Add translated source images alongside stylized images, alternating inputs, with adversarial loss (ablation) | Stage 1 segmentation training · GTA5 → Cityscapes | mIoU: 43.2 -> 44.6 | Helped |
| Add one iteration of target self-training (ablation) | Stage 1 model · GTA5 → Cityscapes | mIoU: 44.6 -> 48.6 | Helped |
| Add a second iteration of target self-training (ablation) | Stage 2 model · GTA5 → Cityscapes | mIoU: 48.6 -> 50.2 | Helped |
| Add a third self-training iteration (ablation) | Stage 2 model · GTA5 → Cityscapes | mIoU: 50.2 -> 50.2 | No clear effect |
| Train with stylized rather than original synthetic images (ablation) | Semantic segmentation model · SYNTHIA → Cityscapes, road | IoU: 79.2 -> 87.7 | Helped |
| Train with stylized rather than original synthetic images (ablation) | Semantic segmentation model · SYNTHIA → Cityscapes, traffic light | IoU: 9.9 -> 1.0 | Hurt |
| Use stylized-source training rather than AdaptSegNet trained on original synthetic images (ablation) | Semantic segmentation model · Cityscapes val, Gaussian noise | score: 22.2 -> 35.1 | Helped |
| Use stylized-source training rather than AdaptSegNet trained on original synthetic images (ablation) | Semantic segmentation model · Cityscapes val, Impulse noise | score: 20.9 -> 32.6 | Helped |
| Use stylized-source training rather than AdaptSegNet trained on original synthetic images (ablation) | Semantic segmentation model · Cityscapes val, Shot noise | score: 24.9 -> 38.2 | Helped |
Closest papers by meaning
- All About Structure: Adapting Structural Information Across Domains for Boosting Semantic Segmentation (CVPR 2019 · similarity 0.91)
- Not All Areas Are Equal: Transfer Learning for Semantic Segmentation via Hierarchical Region Selection (CVPR 2019 · similarity 0.90)
- Unsupervised Domain Adaptation for Semantic Segmentation by Content Transfer (AAAI 2021 · similarity 0.89)
- Addressing Domain Gap via Content Invariant Representation for Semantic Segmentation (AAAI 2021 · similarity 0.89)
- Exploring Semantic Consistency and Style Diversity for Domain Generalized Semantic Segmentation (AAAI 2025 · similarity 0.87)
- Learning Frequency-Adapted Vision Foundation Model for Domain Generalized Semantic Segmentation (NeurIPS 2024 · 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.