Research map / Label-efficient and robust learning / Domain adaptation and generalization
Geometric Unsupervised Domain Adaptation for Semantic Segmentation
Vitor Guizilini, Jie Li, Rareș Ambruș, Adrien Gaidon
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 geometric self-supervision (photometric loss on target videos) to pseudo-label training (ablation) | GUDA · Cityscapes SYNTHIA | mIoU: 46.8 -> 51.0 | Improved |
| add geometric self-supervision to virtual supervision (ablation) | GUDA · Cityscapes SYNTHIA | mIoU: 33.8 -> 44.5 | Improved |
| add surface normal supervision on virtual samples (ablation) | GUDA · Cityscapes SYNTHIA | mIoU: 43.4 -> 44.5 | Improved |
| use 30 context frames instead of 2 (ablation) | GUDA · Cityscapes SYNTHIA | mIoU: 43.8 -> 44.5 | Improved |
| train DADA with 30 frames instead of 1 (ablation) | DADA · Cityscapes SYNTHIA | mIoU: 42.6 -> 42.4 | No clear effect |
| train at multiple resolutions instead of full resolution only (ablation) | GUDA · Cityscapes SYNTHIA | mIoU: 44.3 -> 44.5 | Improved |
| add domain adversarial DANN loss (ablation) | GUDA · Cityscapes SYNTHIA | mIoU: 44.5 -> 44.8 | Improved |
| add USAMR pseudo-label supervision (ablation) | GUDA · Cityscapes SYNTHIA | mIoU: 44.5 -> 51.0 | Improved |
| raise pseudo-label loss weight beyond 0.01 (ablation) | GUDA+PL · Cityscapes SYNTHIA | mIoU | Hurt |
| add partially-supervised photometric loss (ablation) | GUDA (R18) · KITTI depth | Abs.Rel: 0.114 -> 0.109 | Improved |
| add partially-supervised photometric loss (ablation) | GUDA (R18) · DDAD depth | Abs.Rel: 0.166 -> 0.158 | Improved |
| replace fine-tuning with GUDA training | ResNet18 depth network · KITTI depth | Abs.Rel: 0.114 -> 0.109 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SPIGAN — SPIGAN: Privileged Adversarial Learning from Simulation (ICLR 2019) · better on 1 of 1 shared result
- DADA — DADA: Depth-Aware Domain Adaptation in Semantic Segmentation (ICCV 2019) · 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
- Unsupervised Domain Adaptation for Semantic Segmentation via Class-Balanced Self-Training (ECCV 2018 · similarity 0.92)
- Domain Adaptive Semantic Segmentation With Self-Supervised Depth Estimation (ICCV 2021 · similarity 0.91)
- Unsupervised Domain Adaptation for Semantic Segmentation by Content Transfer (AAAI 2021 · similarity 0.91)
- Unsupervised Model Adaptation for Continual Semantic Segmentation (AAAI 2021 · similarity 0.91)
- Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · similarity 0.90)
- Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · 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.
