Research map / Label-efficient and robust learning / Domain adaptation and generalization
Prototypical Cross-Domain Self-Supervised Learning for Few-Shot Unsupervised Domain Adaptation
Xiangyu Yue, Zangwei Zheng, Shanghang Zhang, Yang Gao, Trevor Darrell, Kurt Keutzer, Alberto Sangiovanni Vincentelli
Check your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use PCS instead of the strongest reported baseline | Few-shot domain adaptation · Office, 1-shot, Avg | target accuracy (%): 66.1 -> 76.6 | Improved |
| Use PCS instead of the strongest reported baseline | Few-shot domain adaptation · Office, 3-shots, Avg | target accuracy (%): 80.6 -> 84.0 | Improved |
| Use PCS instead of the strongest reported baseline | Few-shot domain adaptation · Office-Home, 3% labeled source, Avg | target accuracy (%): 56.2 -> 59.7 | Improved |
| Use PCS instead of the strongest reported baseline | Few-shot domain adaptation · Office-Home, 6% labeled source, Avg | target accuracy (%): 60.1 -> 62.6 | Improved |
| Use PCS instead of the strongest reported baseline | Few-shot domain adaptation · VisDA-2017, 0.1% labeled source | target accuracy (%): 69.0 -> 78.0 | Improved |
| Use PCS instead of the strongest reported baseline | Few-shot domain adaptation · VisDA-2017, 1% labeled source | target accuracy (%): 78.3 -> 79.0 | Improved |
| Use PCS instead of the strongest reported baseline | Few-shot domain adaptation · DomainNet, 1-shot, Avg | target accuracy (%): 21.5 -> 34.7 | Improved |
| Use PCS instead of the strongest reported baseline | Few-shot domain adaptation · DomainNet, 3-shots, Avg | target accuracy (%): 42.5 -> 46.1 | Improved |
| Add in-domain prototypical self-supervision (ablation) | Classification-loss-only model · Office, 1-shot, Avg | target accuracy (%): 44.2 -> 52.1 | Improved |
| Add cross-domain instance-prototype self-supervision (ablation) | Model with classification loss and in-domain self-supervision · Office, 1-shot, Avg | target accuracy (%): 52.1 -> 60.1 | Improved |
| Add mutual information maximization (ablation) | Model with in-domain and cross-domain self-supervision · Office, 1-shot, Avg | target accuracy (%): 60.1 -> 64.9 | Improved |
| Add adaptive prototype-classifier update (ablation) | Model with classification, both self-supervision losses, and MIM · Office, 1-shot, Avg | target accuracy (%): 64.9 -> 76.6 | Improved |
Closest papers by meaning
- Mutual Nearest Neighbor Contrast and Hybrid Prototype Self-Training for Universal Domain Adaptation (AAAI 2022 · similarity 0.89)
- High-Level Semantic Feature Matters Few-Shot Unsupervised Domain Adaptation (AAAI 2023 · similarity 0.89)
- Understanding Cross-Domain Few-Shot Learning Based on Domain Similarity and Few-Shot Difficulty (NeurIPS 2022 · similarity 0.89)
- CDS: Cross-Domain Self-Supervised Pre-Training (ICCV 2021 · similarity 0.89)
- Tackling Dimensional Collapse toward Comprehensive Universal Domain Adaptation (ICML 2025 · similarity 0.88)
- Universal Semi-Supervised Domain Adaptation by Mitigating Common-Class Bias (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
