Research map / Label-efficient and robust learning / Domain adaptation and generalization
Foundation-model domain generalization
Uses vision-language models and unsupervised or unknown-domain strategies for generalization.
56 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2020–2026) · an approach in Domain adaptation and generalization, keyword cluster “domain generalization · dg · unseen”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 2 of 6 tracked benchmarks.
26 papers in 2023–24 and 16 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 7.6% to 5.0%.
Open questions the data points to
- OfficeHome (accuracy): best went from 86.13 (up to 2024) to 87.65 in 2025 — Domain Generalization in CLIP via Learning with Diverse Text Prompts (CVPR 2025 · 12 citations)
- DomainNet (accuracy): best went from 62.1 (up to 2024) to 63.11 in 2025 — Domain Generalization in CLIP via Learning with Diverse Text Prompts (CVPR 2025 · 12 citations)
No longer reported by papers here: I&C&M to O (HTER (%), last in 2023); I&C&M to O (AUC (%), 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
- 2020 · Single-Side Domain Generalization for Face Anti-Spoofing (CVPR 2020 · 315 citations)
- 2021 · FSDR: Frequency Space Domain Randomization for Domain Generalization (CVPR 2021 · 286 citations)
- 2022 · Causality Inspired Representation Learning for Domain Generalization (CVPR 2022 · 223 citations)
- 2023 · Sharpness-Aware Gradient Matching for Domain Generalization (CVPR 2023 · 118 citations)
- 2024 · Gradient Alignment for Cross-Domain Face Anti-Spoofing (CVPR 2024 · 41 citations)
- 2025 · Domain Generalization in CLIP via Learning with Diverse Text Prompts (CVPR 2025 · 12 citations)
- 2026 · Balancing Multimodal Domain Generalization via Gradient Modulation and Projection (AAAI 2026 · 1 citation)
Earliest papers here
- Single-Side Domain Generalization for Face Anti-Spoofing (CVPR 2020 · 315 citations)
- FSDR: Frequency Space Domain Randomization for Domain Generalization (CVPR 2021 · 286 citations)
- Progressive Domain Expansion Network for Single Domain Generalization (CVPR 2021 · 195 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.
PACS · accuracy (higher is better)
12 papers, 2021–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 86.2 | Exploiting Domain-Specific Features to Enhance Domain Generalization (NeurIPS 2021 · 70 citations) |
| 2022 | 88.4 | DNA: Domain Generalization with Diversified Neural Averaging (ICML 2022) |
| 2023 | 88.5 | POEM: Polarization of Embeddings for Domain-Invariant Representations (AAAI 2023 · 13 citations) |
| 2024 | 97.45 | Disentangled Prompt Representation for Domain Generalization (CVPR 2024 · 28 citations) |
OfficeHome · accuracy (higher is better)
11 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 69.2 | Exploiting Domain-Specific Features to Enhance Domain Generalization (NeurIPS 2021 · 70 citations) |
| 2022 | 71.2 | DNA: Domain Generalization with Diversified Neural Averaging (ICML 2022) |
| 2023 | 71.7 | POEM: Polarization of Embeddings for Domain-Invariant Representations (AAAI 2023 · 13 citations) |
| 2024 | 86.13 | Disentangled Prompt Representation for Domain Generalization (CVPR 2024 · 28 citations) |
| 2025 | 87.65 | Domain Generalization in CLIP via Learning with Diverse Text Prompts (CVPR 2025 · 12 citations) |
DomainNet · accuracy (higher is better)
10 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 42.8 | Exploiting Domain-Specific Features to Enhance Domain Generalization (NeurIPS 2021 · 70 citations) |
| 2022 | 47.2 | DNA: Domain Generalization with Diversified Neural Averaging (ICML 2022) |
| 2024 | 62.1 | Disentangled Prompt Representation for Domain Generalization (CVPR 2024 · 28 citations) |
| 2025 | 63.11 | Domain Generalization in CLIP via Learning with Diverse Text Prompts (CVPR 2025 · 12 citations) |
VLCS · accuracy (higher is better)
8 papers, 2021–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 79 | Exploiting Domain-Specific Features to Enhance Domain Generalization (NeurIPS 2021 · 70 citations) |
| 2023 | 80 | Sharpness-Aware Gradient Matching for Domain Generalization (CVPR 2023 · 118 citations) |
| 2024 | 86.43 | Disentangled Prompt Representation for Domain Generalization (CVPR 2024 · 28 citations) |
I&C&M to O · HTER (%) (lower is better)
4 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 15.61 | Single-Side Domain Generalization for Face Anti-Spoofing (CVPR 2020 · 315 citations) |
| 2021 | 15.27 | Generalizable Representation Learning for Mixture Domain Face Anti-Spoofing (AAAI 2021 · 125 citations) |
| 2023 | 8.86 | Instance-Aware Domain Generalization for Face Anti-Spoofing (CVPR 2023 · 101 citations) |
I&C&M to O · AUC (%) (higher is better)
4 papers, 2020–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 91.54 | Single-Side Domain Generalization for Face Anti-Spoofing (CVPR 2020 · 315 citations) |
| 2023 | 97.14 | Instance-Aware Domain Generalization for Face Anti-Spoofing (CVPR 2023 · 101 citations) |
What has been tried
47 papers here report 555 measured changes (a component added, removed or swapped, compared with a baseline): helped 77%, no clear effect 3%, hurt 16%, 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 |
|---|---|---|---|---|
| Replace GAC-FAS with the gradient regularizer ⟨∇Lpi(Si), ∇L(S)⟩ | ICM → O · HTER | 9.89 -> 11.35 | Hurt | Gradient Alignment for Cross-Domain Face Anti-Spoofing (CVPR 2024 · 41 citations) |
| add AMB to the encoder-decoder baseline | A → B · MAE | 21.6 -> 18.9 | Helped | Single Domain Generalization for Crowd Counting (CVPR 2024 · 37 citations) |
| add GPT-Assist descriptions to CoOp prompts | VLCS · accuracy | 77.6 -> 79.0 | Helped | Disentangled Prompt Representation for Domain Generalization (CVPR 2024 · 28 citations) |
| use semantics transformations instead of none of the three components | Digits (MNIST source; four target domains, Avg.) · accuracy (%) | 59.3±1.5 -> 77.0±0.4 | Helped | AdvST: Revisiting Data Augmentations for Single Domain Generalization (AAAI 2024 · 25 citations) |
| add instance perturbation | OfficeHome (average) · H-score | 53.92 -> 54.12 | Helped | PracticalDG: Perturbation Distillation on Vision-Language Models for Hybrid Domain Generalization (CVPR 2024 · 16 citations) |
| add domain-sensitive channel suppression with entropy-based prompt selection | OH, TI, VLCS, PACS Avg · accuracy | 81.46 -> 82.28 | Helped | Domain Generalization in CLIP via Learning with Diverse Text Prompts (CVPR 2025 · 12 citations) |
| freeze early blocks | DGSS average · mIoU | 64.4 -> 65.0 (↑0.6) | Helped | SoMA: Singular Value Decomposed Minor Components Adaptation for Domain Generalizable Representation Learning (CVPR 2025 · 9 citations) |
| Add SAM | PACS · accuracy | 81.32 -> 82.31 | Helped | Sharpness-Aware Model-Agnostic Long-Tailed Domain Generalization (AAAI 2024 · 6 citations) |
| Add interpolated domains without extrapolated domains | PACS · accuracy (Avg) | 82.75 -> 84.48 | Helped | Mixup-Induced Domain Extrapolation for Domain Generalization (AAAI 2024 · 5 citations) |
| Add meta prompt initialization instead of zero initialization | Office-Home attention-based generalization (avg.) · accuracy | 65.85 -> 77.06 | Helped | A2XP: Towards Private Domain Generalization (CVPR 2024 · 5 citations) |
| Add SSH | CUB-Paintings: C → P · top-1 accuracy | 65.01 -> 66.42 | Helped | Learning Fine-grained Domain Generalization via Hyperbolic State Space Hallucination (AAAI 2025 · 4 citations) |
| Add DADC and circuit-domain-aware classification | Mfs2 and Resub, Log2, Hyp, Multiplier, and Sin · top 50% accuracy | See Ablation Study table | Helped | A Circuit Domain Generalization Framework for Efficient Logic Synthesis in Chip Design (ICML 2024 · 3 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 |
|---|---|---|---|---|
| ERM | DomainNet | accuracy | 40.9 – 44.1 | 3 papers, e.g. ICML 2022 and AAAI 2023 |
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
- PACS · 3 of 11 papers
- VLCS · 3 of 11 papers
- TerraIncognita · 3 of 11 papers
- OfficeHome · 3 of 11 papers
- DomainNet · 2 of 11 papers
Baselines
- ERM · 4 of 11 papers
- SWAD · 4 of 11 papers
- MIRO · 3 of 11 papers
- SAGM · 3 of 11 papers
- CORAL · 3 of 11 papers
- MixStyle · 2 of 11 papers
- TriD · 2 of 11 papers
- LoRA · 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 42 papers; from OpenAlex and the papers' affiliations).
- University of Science and Technology of China · 3 papers
- The Hong Kong Polytechnic University · 2 papers
- Peking University · 2 papers
- Sungkyunkwan University · 2 papers
- Southeast University · 2 papers
- Nanjing University · 2 papers
- Beijing Institute of Technology · 2 papers
- Tencent · 2 papers
- University of Seoul · 2 papers
- Alibaba Group · 2 papers
Use it
Download BibTeX of 36 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
- Optimization-based domain generalization (101 papers)
- Augmentation and mixture-of-experts generalization (51 papers)
- Generalizable person re-identification (11 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)
- 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).