Research map / Label-efficient and robust learning / Domain adaptation and generalization
Learn from the Learnt: Source-Free Active Domain Adaptation via Contrastive Sampling and Visual Persistence
Mengyao Lyu, Tianxiang Hao, Xinhao Xu, Hui Chen, Zijia Lin, Jungong Han, Guiguang Ding
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 |
|---|---|---|---|
| Use contrastive active sampling (CAS) margin instead of vanilla BvSB (ablation) | LFTL · VisDA-C 1% budget | accuracy (%): 85.1 -> 86.7 | Improved |
| Add class-level transferability term to the sampling criterion (ablation) | LFTL CAS · VisDA-C 1% budget | accuracy (%): 86.7 -> 86.9 | Improved |
| Add anchor-based clustering loss L_ac (ablation) | LFTL · VisDA-C 1% budget | accuracy (%): 86.9 -> 87.0 | Improved |
| Replace L_ac with visual-persistence-guided L_vpa (EMA anchor features) (ablation) | LFTL · Office-31 5% budget | accuracy (%): 92.0 -> 93.5 | Improved |
| Replace L_ac with L_vpa (ablation) | LFTL · VisDA-C 1% budget | accuracy (%): 87.0 -> 87.4 | Improved |
| Use CAS instead of other active learning criteria (Random, entropy, K-means, etc.) (ablation) | LFTL · VisDA-C 1% budget, 10 rounds | accuracy (%): 85.1 (BvSB) -> 87.4 | Improved |
| Sample with entropy-min instead of random (ablation) | Active learning baselines · VisDA-C 1% budget, 10 rounds | accuracy (%): 82.7 -> 63.9 | Hurt |
| Plug CAS sampling into SHOT with 1% labels | SHOT · VisDA-C | accuracy (%): 85.5 -> 86.4 | Improved |
| Plug CAS sampling into SHOT++ with 1% labels | SHOT++ · VisDA-C | accuracy (%): 87.3 -> 87.5 | No clear effect |
| Remove source data loss from LADA (SF-LADA) instead of using hypothesis transfer (ablation) | LADA · source-free comparison | accuracy | Hurt |
| Add source cross-entropy loss to LFTL (S-LFTL) (ablation) | LFTL · Office datasets | accuracy | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- MHPL — MHPL: Minimum Happy Points Learning for Active Source Free Domain Adaptation (CVPR 2023) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Revisiting Source-Free Domain Adaptation: Insights into Representativeness, Generalization, and Variety (CVPR 2025 · similarity 0.91)
- Revisiting the Domain Shift and Sample Uncertainty in Multi-source Active Domain Transfer (CVPR 2024 · similarity 0.91)
- Category-Aware Active Domain Adaptation (ICML 2024 · similarity 0.90)
- Diffusion-Driven Progressive Target Manipulation for Source-Free Domain Adaptation (NeurIPS 2025 · similarity 0.90)
- Adapting to Distribution Shift by Visual Domain Prompt Generation (ICLR 2024 · similarity 0.89)
- Gradient-Based Sample Selection for Black-Box Universal Domain Adaptation (AAAI 2025 · similarity 0.89)
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.
