Research map / Continual learning / Domain and prompt-based
Continual test-time adaptation
Adapts models online to shifting target domains using adapters, prompts and knowledge fusion.
25 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2022–2026) · an approach in Domain and prompt-based, keyword cluster “domains · time adaptation · test time”
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.
11 papers in 2023–24 and 12 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 25.0% to 24.5%.
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
- 2022 · Continual Test-Time Domain Adaptation (CVPR 2022 · 484 citations)
- 2023 · Decorate the Newcomers: Visual Domain Prompt for Continual Test Time Adaptation (AAAI 2023 · 80 citations)
- 2024 · Each Test Image Deserves A Specific Prompt: Continual Test-Time Adaptation for 2D Medical Image Segmentation (CVPR 2024 · 44 citations)
- 2025 · Maintaining Consistent Inter-Class Topology in Continual Test-Time Adaptation (CVPR 2025 · 9 citations)
- 2026 · Shared & Domain Self-Adaptive Experts with Frequency-Aware Discrimination for Continual Test-Time Adaptation (AAAI 2026 · 1 citation)
Earliest papers here
- Continual Test-Time Domain Adaptation (CVPR 2022 · 484 citations)
- Efficient Test-Time Model Adaptation without Forgetting (ICML 2022 · 65 citations)
- Decorate the Newcomers: Visual Domain Prompt for Continual Test Time Adaptation (AAAI 2023 · 80 citations)
What has been tried
17 papers here report 223 measured changes (a component added, removed or swapped, compared with a baseline): helped 74%, no clear effect 2%, hurt 23%, mixed 0%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add a low-frequency prompt to the source-only model | OD/OC segmentation, average across domains A–E · DSC | 65.86 -> 66.90 | Helped | Each Test Image Deserves A Specific Prompt: Continual Test-Time Adaptation for 2D Medical Image Segmentation (CVPR 2024 · 44 citations) |
| add random masking and consistency constraints | CIFAR10-to-CIFAR10C · classification error rate (%) | 28.2 -> 17.1 | Helped | Continual-MAE: Adaptive Distribution Masked Autoencoders for Continual Test-Time Adaptation (CVPR 2024 · 21 citations) |
| Add Lalign only | CIFAR10-C · error rate (%) | 16.05 -> 15.92 | Helped | Maintaining Consistent Inter-Class Topology in Continual Test-Time Adaptation (CVPR 2025 · 9 citations) |
| add self-adaptive threshold label selection and discard unreliable labels | CIFAR10-to-CIFAR10C · mean error rate (%) | 20.4 -> 17.7 | Helped | A Versatile Framework for Continual Test-Time Domain Adaptation: Balancing Discriminability and Generalizability (CVPR 2024 · 6 citations) |
| Add APM to the baseline | ModelNet40 reconstruction · CD (× 10⁻³) | 78.9 -> 23.5 | Helped | PCoTTA: Continual Test-Time Adaptation for Multi-Task Point Cloud Understanding (NeurIPS 2024 · 4 citations) |
| Add a Model Reservoir | CIFAR-100-C, recurring CSC, visit 20 · error (%) | +1.9% | Helped | ReservoirTTA: Prolonged Test-time Adaptation for Evolving and Recurring Domains (NeurIPS 2025 · 2 citations) |
| Add source domain augmentation (SDA) | CDS-Easy · mIoU | 58.6 -> 60.2 | Helped | BECoTTA: Input-dependent Online Blending of Experts for Continual Test-time Adaptation (ICML 2024 · 1 citation) |
| Replace BN-statistic KL initialization with prediction-entropy initialization | CIFAR-100C (15-domain mean) · error rate (%) | 37.1 -> 53.6 | Hurt | CONTRAST: Continual Multi-source Adaptation to Dynamic Distributions (NeurIPS 2024 · 1 citation) |
| Add Prompt Coreset to VPA and Dynamic Update | ImageNet-to-ImageNet-C CSC · error rate | 50.8 -> 39.9 | Helped | DPCore: Dynamic Prompt Coreset for Continual Test-Time Adaptation (ICML 2025 · 0 citations) |
| Remove masked consistency loss (MCL) | CIFARC · classification error rate | Hurt | Ranked Entropy Minimization for Continual Test-Time Adaptation (ICML 2025 · 0 citations) | |
| add domain prompts with KFI and KFU | ImageNet-to-ImageNet-C · mean error rate (%) | 55.8 -> 39.5 | Helped | Class-aware Domain Knowledge Fusion and Fission for Continual Test-Time Adaptation (NeurIPS 2025 · 0 citations) |
| Add high-rank ViDA | ImageNet-to-ImageNet-C · mean classification error rate (%) | 55.8 -> 50.7 | Helped | ViDA: Homeostatic Visual Domain Adapter for Continual Test Time Adaptation (ICLR 2024) |
What reviewers will expect
Datasets and baselines named in the 7 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR10-C · 4 of 7 papers
- CIFAR100-C · 4 of 7 papers
- ImageNet-C · 4 of 7 papers
Baselines
- Tent · 5 of 7 papers
- CoTTA · 5 of 7 papers
- SAR · 4 of 7 papers
- EATA · 3 of 7 papers
- RDumb · 3 of 7 papers
- VDP · 3 of 7 papers
- RoTTA · 3 of 7 papers
- ViDA · 3 of 7 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 23 papers; from OpenAlex and the papers' affiliations).
- Peking University · 4 papers
- AI2Robotics · 2 papers
- Korea University · 2 papers
- Dolby · 2 papers
- Amazon · 2 papers
Use it
Download BibTeX of 25 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-incremental learning with memory (49 papers)
Elsewhere in the topic
- Prompt-based continual learning (26 papers)
- Lifelong person re-identification (14 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).