Research map / Label-efficient and robust learning / Domain adaptation and generalization
Learning to Adapt Structured Output Space for Semantic Segmentation
Read the paperCheck 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 |
|---|---|---|---|
| Adapt segmentation outputs rather than intermediate features (ablation) | ResNet-101 segmentation model · GTA5 → Cityscapes | mIoU: 39.3 -> 41.4 | Helped |
| Add a second adversarial output-adaptation level (ablation) | Ours (single-level) · GTA5 → Cityscapes | mIoU: 41.4 -> 42.4 | Helped |
| Adapt segmentation outputs rather than intermediate features (ablation) | ResNet-101 segmentation model · SYNTHIA → Cityscapes | mIoU: 40.8 -> 45.9 | Helped |
| Add a second adversarial output-adaptation level (ablation) | Ours (single-level) · SYNTHIA → Cityscapes | mIoU: 45.9 -> 46.7 | Helped |
| Adapt segmentation outputs rather than intermediate features (ablation) | ResNet-101 segmentation model · Cityscapes → Cross-City | mIoU: 51.4 -> 53.8 (Rome); 49.7 -> 51.6 (Rio); 48.0 -> 49.9 (Tokyo); 46.6 -> 49.1 (Taipei) | Mixed |
| Adapt GTA5-trained model to target segmentation outputs | ResNet-101 segmentation model · GTA5 → Cityscapes | mIoU: 36.6 -> 42.4 | Helped |
| Adapt SYNTHIA-trained model to target segmentation outputs | ResNet-101 segmentation model · SYNTHIA → Cityscapes | mIoU: 38.6 -> 46.7 | Helped |
| Set λadv to 0.001 rather than 0.0005 for feature adaptation (ablation) | Feature adaptation model · GTA5 → Cityscapes | mIoU: 35.3 -> 39.3 | Helped |
| Set λadv to 0.004 rather than 0.001 for feature adaptation (ablation) | Feature adaptation model · GTA5 → Cityscapes | mIoU: 39.3 -> 32.8 | Hurt |
| Set λadv to 0.0005 rather than 0.001 for output-space adaptation (ablation) | Output-space adaptation model · GTA5 → Cityscapes | mIoU: 41.4 -> 40.2 | Hurt |
| Set λadv to 0.004 rather than 0.001 for output-space adaptation (ablation) | Output-space adaptation model · GTA5 → Cityscapes | mIoU: 41.4 -> 40.1 | Hurt |
Closest papers by meaning
- Joint Adversarial Learning for Domain Adaptation in Semantic Segmentation (AAAI 2020 · similarity 0.91)
- An Adversarial Perturbation Oriented Domain Adaptation Approach for Semantic Segmentation (AAAI 2020 · similarity 0.91)
- Multi-source Domain Adaptation for Semantic Segmentation (NeurIPS 2019 · similarity 0.91)
- Taking a Closer Look at Domain Shift: Category-Level Adversaries for Semantics Consistent Domain Adaptation (CVPR 2019 · similarity 0.90)
- Learning From Synthetic Data: Addressing Domain Shift for Semantic Segmentation (CVPR 2018 · similarity 0.90)
- ADVENT: Adversarial Entropy Minimization for Domain Adaptation in Semantic Segmentation (CVPR 2019 · similarity 0.89)
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.