Research map / Label-efficient and robust learning / Domain adaptation and generalization
Vision-Language Model Guided Source-Free Domain Adaptation via Optimal Transport
Shuo Han, Xu Tang, Jingjing Ma, Xiangrong Zhang
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 |
|---|---|---|---|
| add OT alignment loss L_Align with VLM-guided plan (ablation) | VSFOT · Office-Home | avg top-1 accuracy: 59.92 -> 81.46 | Improved |
| add model-to-VLM distillation L_VLM (ablation) | VSFOT · Office-Home | avg top-1 accuracy: 81.46 -> 84.17 | Improved |
| add model-to-VLM distillation L_VLM (ablation) | VSFOT · DomainNet-126 | avg top-1 accuracy: 81.19 -> 82.44 | Improved |
| add contrastive loss and information maximization (ablation) | VSFOT · DomainNet-126 | avg top-1 accuracy: 82.44 -> 82.63 | No clear effect |
| add contrastive loss and information maximization (ablation) | VSFOT · Office-Home | avg top-1 accuracy: 84.17 -> 85.34 | Improved |
| estimate class marginal from VLM predictions instead of uniform (ablation) | VSFOT · Office-Home | avg top-1 accuracy: 73.62 -> 85.34 | Improved |
| estimate class marginal from VLM predictions instead of uniform (ablation) | VSFOT · DomainNet-126 | avg top-1 accuracy: 64.34 -> 82.63 | Improved |
| add VLM prior to the OT cost (ablation) | VSFOT · Office-Home | avg top-1 accuracy: 74.13 -> 85.34 | Improved |
| add VLM prior to the OT cost (ablation) | VSFOT · DomainNet-126 | avg top-1 accuracy: 64.19 -> 82.63 | Improved |
| replace OT alignment with KL soft-label alignment (ablation) | VSFOT · Office-Home | avg top-1 accuracy: 85.34 -> 64.90 | Hurt |
| replace L_VLM with soft KL distillation (ablation) | VSFOT · Office-Home | avg top-1 accuracy: 85.34 -> 75.41 | Hurt |
| replace L_VLM with hard-label pseudo-label supervision (ablation) | VSFOT · Office-Home | avg top-1 accuracy: 85.34 -> 85.09 | Hurt |
Closest papers by meaning
- Source-Free Domain Adaptation with Vision-Language Prior (IJCV 2026 · similarity 0.94)
- Source-Free Domain Adaptation with Frozen Multimodal Foundation Model (CVPR 2024 · similarity 0.91)
- Learn from the Learnt: Source-Free Active Domain Adaptation via Contrastive Sampling and Visual Persistence (ECCV 2024 · similarity 0.89)
- Prompt-Based Distribution Alignment for Unsupervised Domain Adaptation (AAAI 2024 · similarity 0.89)
- Unified Language-driven Zero-shot Domain Adaptation (CVPR 2024 · similarity 0.88)
- Proxy Denoising for Source-Free Domain Adaptation (ICLR 2025 · 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.
