Research map / Label-efficient and robust learning / Domain adaptation and generalization
Source-Free Domain Adaptation with Vision-Language Prior
Song Tang, Yunxiang Bai, Wenxin X. Su, Mao Ye, Jianwei Zhang, Xiatian Zhu
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 L_cac when training with L_S-I, L_pc and L_rc (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 80.7 -> 89.3 | Improved |
| add L_pc to the other losses (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 81.7 -> 89.3 | Improved |
| add L_rc to the other losses (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 88.6 -> 89.3 | Improved |
| add L_S-I to the other losses (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 87.7 -> 89.3 | Improved |
| optimize mutual information instead of KL divergence (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 82.8 -> 89.3 | Improved |
| use referenced-entropy hard data sampling instead of random selection (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 87.8 -> 89.3 | Improved |
| use referenced uncertainty instead of conventional entropy (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 87.9 -> 89.3 | Improved |
| use prompt learning to customize CLIP instead of fixed template (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 87.7 -> 89.3 | Improved |
| use weighted combination of target and ViL predictions vs target prediction only (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 83.3 -> 89.3 | Improved |
| use weighted combination vs ViL prediction only (ablation) | DIFO++ · Closed-set ablation (avg of three datasets) | average accuracy: 87.6 -> 89.3 | Improved |
| add gap region-driven transfer on top of DIFO | DIFO++ · Office-Home closed-set | average accuracy: 83.1 -> 84.5 | Improved |
| add DIFO++ to continual SFDA vs DIFO | DIFO++ · Office-Home continual SFDA | accuracy: 51 out of 64 cases | Improved |
Closest papers by meaning
- Source-Free Domain Adaptation with Frozen Multimodal Foundation Model (CVPR 2024 · similarity 0.96)
- Vision-Language Model Guided Source-Free Domain Adaptation via Optimal Transport (CVPR 2026 · similarity 0.94)
- Source-Free Domain Adaptation Guided by Vision and Vision-Language Pre-training (IJCV 2024 · similarity 0.91)
- Proxy Denoising for Source-Free Domain Adaptation (ICLR 2025 · similarity 0.90)
- Unified Language-driven Zero-shot Domain Adaptation (CVPR 2024 · similarity 0.88)
- Split to Merge: Unifying Separated Modalities for Unsupervised Domain Adaptation (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.
