Research map / Label-efficient and robust learning / Domain adaptation and generalization
Source-Free Domain Adaptation Guided by Vision and Vision-Language Pre-training
Wenyu Zhang, Li Shen, Chuan-Sheng Foo
Read the paperCheck 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 |
|---|---|---|---|
| co-learn with ImageNet Swin-B instead of the same ResNet-50 | Co-learn · Office-Home | accuracy: 70.0 -> 83.5 | Improved |
| add Co-learn pseudolabels (Swin-B) to existing SFDA methods | SHOT, SHOT++, NRC, AaD · Office-Home | accuracy: 3.6-9.4% | Improved |
| add CLIP zero-shot text classifier guidance (Co-learn++) over Co-learn | Co-learn (CLIP) · DomainNet | accuracy: 80.3 -> 90.5 | Improved |
| co-learn with the same ImageNet network used to initialize the source model | SHOT, SHOT++ · Office-31 | accuracy: 88.7 -> 88.8 | No clear effect |
| co-learn with ResNet-50 on NRC and AaD (Office-31) | NRC, AaD · Office-31 | accuracy: 0.8% and 0.6% | Improved |
| MatchOrConf pseudolabeling rather than SelfConf (ablation) | Co-learn · Office-Home (A->C,A->P,A->R) | accuracy: 67.0 -> 77.7 | Improved |
| MatchOrConf instead of MatchAndConf (ablation) | Co-learn · Office-Home (A->C,A->P,A->R) | accuracy: 75.4 -> 77.7 | Improved |
| fine-tune the feature extractor of the pre-trained branch (ablation) | Co-learn · Office-Home (A->C,A->P,A->R) | accuracy: 76.6 -> 76.2 | Hurt |
| fine-tune only the weighted NCC classifier in the pre-trained branch (ablation) | Co-learn · Office-Home (A->C,A->P,A->R) | accuracy: 76.6 -> 77.7 | Improved |
| lower confidence threshold gamma from 0.5 to 0.1 (ablation) | Co-learn · VisDA-C | accuracy: 86.4 -> 87.1 | Improved |
| lower confidence threshold gamma from 0.5 to 0.1 (ablation) | Co-learn · Office-Home | accuracy: 77.4 -> 76.1 | Hurt |
| strong instead of weak CLIP zero-shot guidance (ablation) | Co-learn++ · Office-Home | accuracy: 87.1 -> 83.7 | Hurt |
Closest papers by meaning
- Rethinking the Role of Pre-Trained Networks in Source-Free Domain Adaptation (ICCV 2023 · similarity 0.93)
- Source-Free Domain Adaptation with Frozen Multimodal Foundation Model (CVPR 2024 · similarity 0.91)
- Understanding and Improving Source-free Domain Adaptation from a Theoretical Perspective (CVPR 2024 · similarity 0.91)
- Revisiting Source-Free Domain Adaptation: Insights into Representativeness, Generalization, and Variety (CVPR 2025 · similarity 0.89)
- Proxy Denoising for Source-Free Domain Adaptation (ICLR 2025 · similarity 0.88)
- SF(DA)$^2$: Source-free Domain Adaptation Through the Lens of Data Augmentation (ICLR 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.
