Research map / Label-efficient and robust learning / Domain adaptation and generalization
Adaptation for semantic segmentation
Adapts segmentation across domains with open-set, contextual and coarse-to-fine alignment.
81 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Domain adaptation and generalization, keyword cluster “segmentation · semantic · domain adaptive”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 6 tracked benchmarks.
17 papers in 2023–24 and 11 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 5.0% to 3.4%.
Open questions the data points to
- SYNTHIA → Cityscapes (mIoU): best went from 67.7 (up to 2024) to 76.7 in 2025 — MaskTwins: Dual-form Complementary Masking for Domain-Adaptive Image Segmentation (ICML 2025 · 1 citation)
- Papers report Source only on GTA5→Cityscapes (mIoU) anywhere from 24.9 to 37.6: comparisons against it depend on whose number you take.
No longer reported by papers here: GTA5 → Cityscapes (mIoU, last in 2024); GTA5→Cityscapes val (mIoU, last in 2024); SYNTHIA -> Cityscapes (16 classes) (mIoU, last in 2023); SYNTHIA → Cityscapes val (mIoU, last in 2024); SYNTHIA -> Cityscapes (13 classes) (mIoU*, last in 2023).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2018 · Learning to Adapt Structured Output Space for Semantic Segmentation (CVPR 2018 · 1,793 citations)
- 2019 · Taking a Closer Look at Domain Shift: Category-Level Adversaries for Semantics Consistent Domain Adaptation (CVPR 2019 · 846 citations)
- 2020 · FDA: Fourier Domain Adaptation for Semantic Segmentation (CVPR 2020 · 1,119 citations)
- 2021 · Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (CVPR 2021 · 541 citations)
- 2022 · DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation (CVPR 2022 · 616 citations)
- 2023 · MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation (CVPR 2023 · 350 citations)
- 2024 · Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · 30 citations)
- 2025 · Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · 9 citations)
- 2026 · PanDA: Unsupervised Domain Adaptation for Multimodal 3D Panoptic Segmentation in Autonomous Driving (CVPR 2026 · 0 citations)
Earliest papers here
- Learning to Adapt Structured Output Space for Semantic Segmentation (CVPR 2018 · 1,793 citations)
- CyCADA: Cycle-Consistent Adversarial Domain Adaptation (ICML 2018 · 627 citations)
- Learning From Synthetic Data: Addressing Domain Shift for Semantic Segmentation (CVPR 2018 · 520 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
GTA5 → Cityscapes · mIoU (higher is better)
21 papers, 2018–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 39.5 | CyCADA: Cycle-Consistent Adversarial Domain Adaptation (ICML 2018 · 627 citations) |
| 2019 | 48.5 | Bidirectional Learning for Domain Adaptation of Semantic Segmentation (CVPR 2019 · 715 citations) |
| 2020 | 50.45 | FDA: Fourier Domain Adaptation for Semantic Segmentation (CVPR 2020 · 1,119 citations) |
| 2021 | 56.1 | Coarse-To-Fine Domain Adaptive Semantic Segmentation With Photometric Alignment and Category-Center Regularization (CVPR 2021 · 56 citations) |
| 2023 | 73.6 | FREDOM: Fairness Domain Adaptation Approach to Semantic Scene Understanding (CVPR 2023 · 29 citations) |
SYNTHIA → Cityscapes · mIoU (higher is better)
15 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 47.8 | Taking a Closer Look at Domain Shift: Category-Level Adversaries for Semantics Consistent Domain Adaptation (CVPR 2019 · 846 citations) |
| 2020 | 49.3 | Learning Texture Invariant Representation for Domain Adaptation of Semantic Segmentation (CVPR 2020 · 278 citations) |
| 2021 | 54.1 | Cross-Domain Grouping and Alignment for Domain Adaptive Semantic Segmentation (AAAI 2021 · 20 citations) |
| 2023 | 67.7 | Continuous Pseudo-Label Rectified Domain Adaptive Semantic Segmentation With Implicit Neural Representations (CVPR 2023 · 26 citations) |
| 2025 | 76.7 | MaskTwins: Dual-form Complementary Masking for Domain-Adaptive Image Segmentation (ICML 2025 · 1 citation) |
GTA5→Cityscapes val · mIoU (higher is better)
10 papers, 2018–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 42.4 | Learning to Adapt Structured Output Space for Semantic Segmentation (CVPR 2018 · 1,793 citations) |
| 2019 | 50.2 | Category Anchor-Guided Unsupervised Domain Adaptation for Semantic Segmentation (NeurIPS 2019 · 93 citations) |
| 2021 | 57.5 | Prototypical Pseudo Label Denoising and Target Structure Learning for Domain Adaptive Semantic Segmentation (CVPR 2021 · 541 citations) |
| 2022 | 68.3 | DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation (CVPR 2022 · 616 citations) |
SYNTHIA -> Cityscapes (16 classes) · mIoU (higher is better)
7 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 39 | Bidirectional Learning for Domain Adaptation of Semantic Segmentation (CVPR 2019 · 715 citations) |
| 2020 | 40.5 | FDA: Fourier Domain Adaptation for Semantic Segmentation (CVPR 2020 · 1,119 citations) |
| 2021 | 48.2 | Coarse-To-Fine Domain Adaptive Semantic Segmentation With Photometric Alignment and Category-Center Regularization (CVPR 2021 · 56 citations) |
| 2022 | 50.2 | Domain-Agnostic Prior for Transfer Semantic Segmentation (CVPR 2022 · 40 citations) |
| 2023 | 53.8 | When Visual Prompt Tuning Meets Source-Free Domain Adaptive Semantic Segmentation (NeurIPS 2023 · 2 citations) |
SYNTHIA → Cityscapes val · mIoU (higher is better)
7 papers, 2018–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 46.7 | Learning to Adapt Structured Output Space for Semantic Segmentation (CVPR 2018 · 1,793 citations) |
| 2021 | 53.9 | Unsupervised Domain Adaptation for Semantic Segmentation by Content Transfer (AAAI 2021 · 43 citations) |
| 2022 | 70.1 | Towards Fewer Annotations: Active Learning via Region Impurity and Prediction Uncertainty for Domain Adaptive Semantic Segmentation (CVPR 2022 · 97 citations) |
SYNTHIA -> Cityscapes (13 classes) · mIoU* (higher is better)
5 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 51.4 | Bidirectional Learning for Domain Adaptation of Semantic Segmentation (CVPR 2019 · 715 citations) |
| 2020 | 52.5 | FDA: Fourier Domain Adaptation for Semantic Segmentation (CVPR 2020 · 1,119 citations) |
| 2021 | 55.5 | Coarse-To-Fine Domain Adaptive Semantic Segmentation With Photometric Alignment and Category-Center Regularization (CVPR 2021 · 56 citations) |
| 2023 | 71 | Weakly-Supervised Domain Adaptive Semantic Segmentation With Prototypical Contrastive Learning (CVPR 2023 · 30 citations) |
What has been tried
77 papers here report 914 measured changes (a component added, removed or swapped, compared with a baseline): helped 81%, no clear effect 3%, hurt 14%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Expand the classifier head from C to C+1 classes | GTA5 → Cityscapes · H-Score | 19.71 -> 43.79 | Helped | Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · 30 citations) |
| Use entropy, image-level loss, or domain distance instead of stability to divide subsets | GTA5 → Cityscapes · mIoU | 55.6 vs 52.9, 51.7, 52.7, 51.6 | Hurt | Stable Neighbor Denoising for Source-free Domain Adaptive Segmentation (CVPR 2024 · 22 citations) |
| Add prototype adaptation loss (Lppa) to prediction supervision | Cityscapes-to-DensePASS · mIoU | 38.65 -> 45.42 | Helped | Semantics Distortion and Style Matter: Towards Source-free UDA for Panoramic Segmentation (CVPR 2024 · 16 citations) |
| add DSPD to the BUS baseline | Pascal-Context → Cityscapes · H-Score | 36.03 -> 38.04 | Helped | Universal Domain Adaptation for Semantic Segmentation (CVPR 2025 · 9 citations) |
| add DyCE to CT + VLP + DLG | GTA5 → Cityscapes, 100 labeled target images · mIoU | 70.3% to 71.1% | Helped | SemiDAViL: Semi-supervised Domain Adaptation with Vision-Language Guidance for Semantic Segmentation (CVPR 2025 · 7 citations) |
| Output-level late fusion (uniform or proximity-weighted) instead of parameter-level LoRA merging | 20-domain leave-one-out benchmark · h-mean mIoU | -1.99 for SemLA late fusion vs SemLA (52.09 vs 54.16) | Hurt | Semantic Library Adaptation: LoRA Retrieval and Fusion for Open-Vocabulary Semantic Segmentation (CVPR 2025 · 5 citations) |
| Add DMC to Density-aware Greedy | GTAV → Cityscapes, 2.2% labels · mIOU | 70.4 -> 71.1 | Helped | Density Matters: Improved Core-Set for Active Domain Adaptive Segmentation (AAAI 2024 · 3 citations) |
| Add MTP to the frozen VFM (without LST) | DG nuScenes:Sing./USA · mIoU (xM) | +1.3% | Helped | UniDSeg: Unified Cross-Domain 3D Semantic Segmentation via Visual Foundation Models Prior (NeurIPS 2024 · 3 citations) |
| Add random masking with consistency learning | SYNTHIA→Cityscapes · mIoU | 53.7 -> 74.3 | Helped | MaskTwins: Dual-form Complementary Masking for Domain-Adaptive Image Segmentation (ICML 2025 · 1 citation) |
| Add cross-view adaptation to prompting | SYNTHIA → UAVID · mIoU | 20.2 -> 51.1 | Helped | EAGLE: Efficient Adaptive Geometry-based Learning in Cross-view Understanding (NeurIPS 2024 · 1 citation) |
| Add Point Ranker to random point selection | USA→Singapore, 3% labels · 2D+3D mIoU (%) | 77.00 -> 78.32 | Helped | Omni-Query Active Learning for Source-Free Domain Adaptive Cross-Modality 3D Semantic Segmentation (AAAI 2025 · 0 citations) |
| Add post-hoc logit adjustment | GTA→CS · mIoU | 68.3 -> 69.2 | Helped | Balanced Learning for Domain Adaptive Semantic Segmentation (ICML 2025 · 0 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| Source only | GTA5→Cityscapes | mIoU | 24.9 – 37.6 | 6 papers, e.g. AAAI 2021 and CVPR 2021 |
| ProDA | GTA5 → Cityscapes val | mIoU | 53.7 – 57.5 | 2 papers, e.g. CVPR 2022 and CVPR 2022 |
What reviewers will expect
Datasets and baselines named in the 8 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Cityscapes · 7 of 8 papers
- SYNTHIA · 5 of 8 papers
- ImageNet-1k · 3 of 8 papers
- BDD100K · 2 of 8 papers
- Mapillary Vistas · 2 of 8 papers
- ACDC · 2 of 8 papers
- GTA5 · 2 of 8 papers
- IDD · 2 of 8 papers
Baselines
- DAFormer · 5 of 8 papers
- MIC · 5 of 8 papers
- HRDA · 4 of 8 papers
- DACS · 3 of 8 papers
- DIGA · 3 of 8 papers
- ProDA · 2 of 8 papers
- CPSL · 2 of 8 papers
- TransDA · 2 of 8 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 28 papers; from OpenAlex and the papers' affiliations).
- ETH Zurich · 3 papers
- University of Science and Technology of China · 3 papers
- Tencent · 2 papers
- Xiamen University · 2 papers
- Hong Kong University of Science and Technology · 2 papers
- AI Thrust · 2 papers
- Max Planck Institute for Informatics · 2 papers
- Huawei · 2 papers
- University of Arkansas · 2 papers
- Xidian University · 2 papers
Use it
Download BibTeX of 37 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Domain-generalized segmentation (56 papers)
- Cross-modality medical segmentation adaptation (20 papers)
Elsewhere in the topic
- Conditional distribution alignment (162 papers)
- Cross-domain few-shot with transformers (116 papers)
- Feature alignment for unsupervised adaptation (105 papers)
- Optimization-based domain generalization (101 papers)
- Adversarial domain adaptation (95 papers)
- Geometric alignment and synthetic-to-real adaptation (93 papers)
- Source-free domain adaptation (92 papers)
- Statistical alignment and hypothesis testing (80 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).