Research map / Label-efficient and robust learning / Domain adaptation and generalization
Geometric alignment and synthetic-to-real adaptation
Aligns geometry and statistics, simplifies images, and adapts for pose or anytime settings.
93 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Domain adaptation and generalization, keyword cluster “domain adaptation · unsupervised domain · uda”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
28 papers in 2023–24 and 26 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 8.2% to 8.2%.
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
- 2016 · Cold-Start Heterogeneous-Device Wireless Localization (AAAI 2016 · 39 citations)
- 2017 · Joint Geometrical and Statistical Alignment for Visual Domain Adaptation (CVPR 2017 · 593 citations)
- 2018 · Image-Image Domain Adaptation With Preserved Self-Similarity and Domain-Dissimilarity for Person Re-Identification (CVPR 2018 · 1,045 citations)
- 2019 · Geometry-Aware Symmetric Domain Adaptation for Monocular Depth Estimation (CVPR 2019 · 199 citations)
- 2020 · Mutual Mean-Teaching: Pseudo Label Refinery for Unsupervised Domain Adaptation on Person Re-identification (ICLR 2020 · 407 citations)
- 2021 · From Synthetic to Real: Unsupervised Domain Adaptation for Animal Pose Estimation (CVPR 2021 · 91 citations)
- 2022 · SHIFT: A Synthetic Driving Dataset for Continuous Multi-Task Domain Adaptation (CVPR 2022 · 192 citations)
- 2023 · Contrastive Mean Teacher for Domain Adaptive Object Detectors (CVPR 2023 · 137 citations)
- 2024 · CAT: Exploiting Inter-Class Dynamics for Domain Adaptive Object Detection (CVPR 2024 · 47 citations)
- 2025 · Bridge Then Begin Anew: Generating Target-Relevant Intermediate Model for Source-Free Visual Emotion Adaptation (AAAI 2025 · 5 citations)
- 2026 · Domain Adaptation Guided Infrared and Visible Image Fusion (AAAI 2026 · 3 citations)
Earliest papers here
- Cold-Start Heterogeneous-Device Wireless Localization (AAAI 2016 · 39 citations)
- Joint Geometrical and Statistical Alignment for Visual Domain Adaptation (CVPR 2017 · 593 citations)
- Image-Image Domain Adaptation With Preserved Self-Similarity and Domain-Dissimilarity for Person Re-Identification (CVPR 2018 · 1,045 citations)
What has been tried
70 papers here report 788 measured changes (a component added, removed or swapped, compared with a baseline): helped 79%, no clear effect 3%, hurt 15%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add CRA to the base AT framework | Cityscapes → Foggy Cityscapes (0.02) · mAP | 49.3 -> 51.0 | Helped | CAT: Exploiting Inter-Class Dynamics for Domain Adaptive Object Detection (CVPR 2024 · 47 citations) |
| Add labeled target-like images to source-only pretraining | Cityscapes → Foggy Cityscapes (pretraining) · mAP | 28.5 -> 42.3 | Helped | Mean Teacher DETR with Masked Feature Alignment: A Robust Domain Adaptive Detection Transformer Framework (AAAI 2024 · 21 citations) |
| Select similar KADID-10k distortion domains instead of training on all domains | LIVEC · SRCC | 0.5646 -> 0.6958 (+13.12%) | Helped | Bridging the Synthetic-to-Authentic Gap: Distortion-Guided Unsupervised Domain Adaptation for Blind Image Quality Assessment (CVPR 2024 · 18 citations) |
| add WAT | DRealSR (Olympus→Panasonic, ×4) · PSNR | 30.64 -> 31.41 | Helped | Uncertainty-Aware Source-Free Adaptive Image Super-Resolution with Wavelet Augmentation Transformer (CVPR 2024 · 10 citations) |
| add clustering | EmoSet → FI; FI → EmoSet; ArtPhoto → FI; FI → ArtPhoto; Emotion6 → FI; FI → Emotion6 · accuracy | 51.59 -> 51.66; 52.63 -> 53.22; 27.40 -> 28.08; 33.95 -> 34.57; 66.81 -> 73.86; 70.42 -> 72.28 | Helped | Bridge Then Begin Anew: Generating Target-Relevant Intermediate Model for Source-Free Visual Emotion Adaptation (AAAI 2025 · 5 citations) |
| Add unlabeled nighttime data | Nighttime dataset · mean MAE | 7.36 -> 6.95 | Helped | NightCC: Nighttime Color Constancy via Adaptive Channel Masking (CVPR 2024 · 5 citations) |
| Add LoRA-Adapters to Conv-Adapters | Five-domain average, 25 target samples · BD-rate | -15.40% -> -16.43% | Helped | Few-Shot Domain Adaptation for Learned Image Compression (AAAI 2025 · 3 citations) |
| Add image-guided domain adaptation using the LR cross-domain direction | RealSR_Canon · PSNR | 24.824 -> 27.014; 2.19 | Helped | IODA: Instance-Guided One-shot Domain Adaptation for Super-Resolution (NeurIPS 2024 · 3 citations) |
| vary clutter model number N across 1, 5, 20, and 35 | OOD-CV bus category · π/6 accuracy | 95.1; 94.4; 95.5; 94.5 | Mixed | Source-Free and Image-Only Unsupervised Domain Adaptation for Category Level Object Pose Estimation (ICLR 2024 · 1 citation) |
| Use augmentation only, without calibration loss | GTAV→Cityscapes · ECE | 5.34 -> 4.30 | Helped | AUGCAL: Improving Sim2Real Adaptation by Uncertainty Calibration on Augmented Synthetic Images (ICLR 2024 · 1 citation) |
| add outlier classes and BNM loss | S→U · accuracy | 89.1 -> 96.5 | Helped | Image-to-video Adaptation with Outlier Modeling and Robust Self-learning (AAAI 2025 · 0 citations) |
| Use RMLP amplitude 0.1 instead of amplitude 5 | OCT retinal layer segmentation, averaged DICE · DICE | 0.97±0.03 -> 0.94±0.12 | Hurt | Randomized-MLP Regularization Improves Domain Adaptation and Interpretability in DINOv2 (NeurIPS 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 11 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Office-Home · 2 of 11 papers
- DIV2K · 2 of 11 papers
Baselines
- Source only · 2 of 11 papers
- DANN · 2 of 11 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 54 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 4 papers
- Sun Yat-sen University · 3 papers
- Huazhong University of Science and Technology · 3 papers
- National University of Singapore · 3 papers
- ShanghaiTech University · 2 papers
- Harbin Institute of Technology · 2 papers
- University of Science and Technology of China · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- Huawei · 2 papers
- ASUS Intelligent Cloud Services · 2 papers
Use it
Download BibTeX of 40 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
- Conditional distribution alignment (162 papers)
- Feature alignment for unsupervised adaptation (105 papers)
- Source-free domain adaptation (92 papers)
- Adversarial domain adaptation (95 papers)
- Statistical alignment and hypothesis testing (80 papers)
Elsewhere in the topic
- Cross-domain few-shot with transformers (116 papers)
- Optimization-based domain generalization (101 papers)
- Adaptation for semantic segmentation (81 papers)
- Statistical transfer learning (79 papers)
- Mixture and distillation for domain shift (78 papers)
- Foundation-model domain generalization (56 papers)
- Domain-generalized segmentation (56 papers)
- Cross-domain text and graph transfer (54 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).