Research map / Label-efficient and robust learning / Domain adaptation and generalization
Cross-domain few-shot with transformers
Adapts token- and adapter-based vision models to cross-domain few-shot learning and detection.
116 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Domain adaptation and generalization, keyword cluster “cross domain · transfer · domain shot”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Few recent reports None of the tracked benchmarks has a report from 2025–26.
26 papers in 2023–24 and 40 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 7.6% to 12.5%.
No longer reported by papers here: DukeMTMC-reID→Market-1501 (mAP, last in 2020); Market-1501→DukeMTMC-reID (mAP, last in 2020); DukeMTMC-reID→Market-1501 (Rank-1, last in 2020); Market-1501→DukeMTMC-reID (Rank-1, last in 2020).
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 · Learning Multi-Domain Convolutional Neural Networks for Visual Tracking (CVPR 2016 · 2,839 citations)
- 2017 · Correlational Gaussian Processes for Cross-Domain Visual Recognition (CVPR 2017 · 38 citations)
- 2018 · Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-Identification (CVPR 2018 · 702 citations)
- 2019 · Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-Identification (CVPR 2019 · 709 citations)
- 2020 · AD-Cluster: Augmented Discriminative Clustering for Domain Adaptive Person Re-Identification (CVPR 2020 · 320 citations)
- 2021 · Few-Shot Image Generation via Cross-Domain Correspondence (CVPR 2021 · 226 citations)
- 2022 · Detecting Deepfakes With Self-Blended Images (CVPR 2022 · 455 citations)
- 2023 · StyleAdv: Meta Style Adversarial Training for Cross-Domain Few-Shot Learning (CVPR 2023 · 91 citations)
- 2024 · Day-Night Cross-domain Vehicle Re-identification (CVPR 2024 · 36 citations)
- 2025 · FAMNet: Frequency-aware Matching Network for Cross-domain Few-shot Medical Image Segmentation (AAAI 2025 · 29 citations)
- 2026 · Wi-CBR: Salient-aware Adaptive WiFi Sensing for Cross-domain Behavior Recognition (AAAI 2026 · 2 citations)
Earliest papers here
- Learning Multi-Domain Convolutional Neural Networks for Visual Tracking (CVPR 2016 · 2,839 citations)
- Correlational Gaussian Processes for Cross-Domain Visual Recognition (CVPR 2017 · 38 citations)
- Unsupervised Adaptive Re-Identification in Open World Dynamic Camera Networks (CVPR 2017 · 33 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.
DukeMTMC-reID→Market-1501 · mAP (higher is better)
5 papers, 2018–2020 · not reported since 2020
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 26.5 | Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-Identification (CVPR 2018 · 702 citations) |
| 2019 | 43 | Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-Identification (CVPR 2019 · 709 citations) |
| 2020 | 68.3 | AD-Cluster: Augmented Discriminative Clustering for Domain Adaptive Person Re-Identification (CVPR 2020 · 320 citations) |
Market-1501→DukeMTMC-reID · mAP (higher is better)
5 papers, 2018–2020 · not reported since 2020
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 23 | Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-Identification (CVPR 2018 · 702 citations) |
| 2019 | 40.4 | Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-Identification (CVPR 2019 · 709 citations) |
| 2020 | 55.8 | Smoothing Adversarial Domain Attack and P-Memory Reconsolidation for Cross-Domain Person Re-Identification (CVPR 2020 · 72 citations) |
DukeMTMC-reID→Market-1501 · Rank-1 (higher is better)
4 papers, 2018–2020 · not reported since 2020
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 58.2 | Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-Identification (CVPR 2018 · 702 citations) |
| 2019 | 75.1 | Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-Identification (CVPR 2019 · 709 citations) |
| 2020 | 83 | Smoothing Adversarial Domain Attack and P-Memory Reconsolidation for Cross-Domain Person Re-Identification (CVPR 2020 · 72 citations) |
Market-1501→DukeMTMC-reID · Rank-1 (higher is better)
4 papers, 2018–2020 · not reported since 2020
| Year | Best so far | Set by |
|---|---|---|
| 2018 | 44.3 | Transferable Joint Attribute-Identity Deep Learning for Unsupervised Person Re-Identification (CVPR 2018 · 702 citations) |
| 2019 | 63.3 | Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-Identification (CVPR 2019 · 709 citations) |
| 2020 | 74.5 | Smoothing Adversarial Domain Attack and P-Memory Reconsolidation for Cross-Domain Person Re-Identification (CVPR 2020 · 72 citations) |
What has been tried
98 papers here report 1,107 measured changes (a component added, removed or swapped, compared with a baseline): helped 78%, no clear effect 2%, hurt 17%, 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 NGS only | DN-348 Day-to-Night · Rank-1 | 0.674 -> 0.690 | Helped | Day-Night Cross-domain Vehicle Re-identification (CVPR 2024 · 36 citations) |
| Remove feature queries from the backbone-plus-query baseline | Domain-1 · DSC | 84.92 -> 86.16 | Hurt | Learning Generalized Medical Image Segmentation from Decoupled Feature Queries (AAAI 2024 · 31 citations) |
| Add CPG to PANet baseline | Abdominal CT → MRI · Dice | 32.46 -> 53.97 | Helped | FAMNet: Frequency-aware Matching Network for Cross-domain Few-shot Medical Image Segmentation (AAAI 2025 · 29 citations) |
| Use FLoR instead of the BN baseline | CDFSL average, 5-way 5-shot (plain prototype classifier) · accuracy | 57.26 ±0.13 -> 61.52 ±0.12 | Helped | Flatten Long-Range Loss Landscapes for Cross-Domain Few-Shot Learning (CVPR 2024 · 28 citations) |
| Add EST as the inference representation instead of voxel grid | N-CALTECH101 · accuracy | 66.77 -> 66.96; 72.68 -> 73.00; 92.35 -> 92.74 | Helped | EventDance: Unsupervised Source-free Cross-modal Adaptation for Event-based Object Recognition (CVPR 2024 · 21 citations) |
| Replace the final FC layer with GPM, without SOT | DE→DM · gaze estimation error (°) | 8.66±0.53 -> 7.87±0.23 | Helped | From Feature to Gaze: A Generalizable Replacement of Linear Layer for Gaze Estimation (CVPR 2024 · 15 citations) |
| Replace domain adversarial loss with patch similarity loss while retaining Mean Teacher labels | Cityscapes → BDD100k · best mAP | 31.8 -> 35.3 | Helped | Large Self-Supervised Models Bridge the Gap in Domain Adaptive Object Detection (CVPR 2025 · 10 citations) |
| add attention abandonment during source-domain training | CropDiseases, 5-way 5-shot · accuracy | 94.24±0.27 -> 95.40±0.33 | Helped | Attention Temperature Matters in ViT-Based Cross-Domain Few-Shot Learning (NeurIPS 2024 · 10 citations) |
| add the normalization adapter | Meta-Dataset SDL, Avg. dissimilar · accuracy | 81.7 -> 85.8 | Helped | Leveraging Normalization Layer in Adapters with Progressive Learning and Adaptive Distillation for Cross-Domain Few-Shot Learning (AAAI 2024 · 9 citations) |
| not load the source-trained CLS token for target-domain evaluation | CropDisease, EuroSAT, ISIC2018, ChestX; 5-way 1-shot · accuracy | +1.32, +0.94, +0.01, +0.06 | Mixed | A Closer Look at the CLS Token for Cross-Domain Few-Shot Learning (NeurIPS 2024 · 9 citations) |
| replace pixel reconstruction with aggregated multi-layer feature targets and use the Lightweight Decoder | Average, 5-way 5-shot · accuracy | 63.58 -> 65.77 | Helped | Reconstruction Target Matters in Masked Image Modeling for Cross-Domain Few-Shot Learning (AAAI 2025 · 8 citations) |
| add agent enhancement | CD-FSS benchmark, 1-shot · mIoU | 60.00 -> 65.45 | Helped | Dual-Agent Optimization framework for Cross-Domain Few-Shot Segmentation (CVPR 2025 · 8 citations) |
What reviewers will expect
Datasets and baselines named in the 29 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- ISIC2018 · 5 of 29 papers
- miniImageNet · 4 of 29 papers
- EuroSAT · 3 of 29 papers
- ChestX · 3 of 29 papers
- PASCAL VOC 2012 · 3 of 29 papers
- SBD · 3 of 29 papers
- Chest X-ray · 3 of 29 papers
- FSS-1000 · 3 of 29 papers
Baselines
- PATNet · 5 of 29 papers
- PMF · 4 of 29 papers
- StyleAdv · 4 of 29 papers
- APSeg · 4 of 29 papers
- ABCDFSS · 4 of 29 papers
- APM · 4 of 29 papers
- PerSAM · 4 of 29 papers
- PANet · 3 of 29 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 66 papers; from OpenAlex and the papers' affiliations).
- Huazhong University of Science and Technology · 9 papers
- Hong Kong University of Science and Technology · 3 papers
- University of Science and Technology of China · 3 papers
- Korea Advanced Institute of Science and Technology · 2 papers
- University of Electronic Science and Technology of China · 2 papers
- Nanjing University of Aeronautics and Astronautics · 2 papers
- Southeast University · 2 papers
- Yonsei University · 2 papers
- Fudan University · 2 papers
- Beihang University · 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
- Mixture and distillation for domain shift (78 papers)
- Statistical transfer learning (79 papers)
- Cross-domain text and graph transfer (54 papers)
Elsewhere in the topic
- Conditional distribution alignment (162 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)
- Adaptation for semantic segmentation (81 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).