Research map / Label-efficient and robust learning / Domain adaptation and generalization
Interpretable Open-Set Domain Adaptation via Angular Margin Separation
Xinhao Li, Jingjing Li, Zhekai Du, Lei Zhu, Wen Li
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 additive angular margin loss in fine-tuning (ablation) | SR-OSDA base · P->R semantic recovery | U (unseen class recovery): 45.6 -> 50.9 | Improved |
| add additive angular margin loss in fine-tuning (ablation) | SR-OSDA base · P->R open-set recognition | OS-unknown (open-set unseen accuracy): 88.1 -> 90.3 | Improved |
| add angular regularization in fine-tuning (ablation) | AMS fine-tuning margin · P->R semantic recovery | S (seen class recovery): 77.2 -> 77.6 | Improved |
| add angular margin in training phase (ablation) | AMS with fine-tuning · P->R semantic recovery | U (unseen class recovery): 50.7 -> 47.3 | Hurt |
| add angular margin in training phase (ablation) | AMS with fine-tuning · P->R open-set recognition | OS* (seen accuracy): 78.3 -> 82.6 | Improved |
| add angular regularization in training with a single unseen class (ablation) | AMS with training margin · P->R semantic recovery | U (unseen class recovery): 47.3 -> 41.6 | Hurt |
| switch classifier to |Cs+K|-way multiple unseen classes without MRC (ablation) | AMS with training margin · P->R semantic recovery | U (unseen class recovery) | No clear effect |
| add maximum-response correspondence (MRC) with multiple unseen clusters (ablation) | AMS with multi-unseen classifier · P->R semantic recovery | U (unseen class recovery): 12.5% | Improved |
| add post-training attribute re-projection (ablation) | AMS · P->R semantic recovery | S (seen class recovery): 81.7 -> 82.8 | Improved |
| add post-training attribute re-projection (ablation) | AMS · P->R semantic recovery | U (unseen class recovery): 58.1 -> 58.1 | No clear effect |
| increase number of unseen clusters K from 1 toward the ground truth (ablation) | AMS · P->R open-set recognition | OS-unknown and U | Improved |
| increase K beyond ground truth (more than three times) (ablation) | AMS · P->R open-set recognition | all metrics | No clear effect |
Closest papers by meaning
- Towards Novel Target Discovery Through Open-Set Domain Adaptation (ICCV 2021 · similarity 0.91)
- Separate to Adapt: Open Set Domain Adaptation via Progressive Separation (CVPR 2019 · similarity 0.91)
- Adjustment and Alignment for Unbiased Open Set Domain Adaptation (CVPR 2023 · similarity 0.90)
- Open-Set Heterogeneous Domain Adaptation: Theoretical Analysis and Algorithm (AAAI 2025 · similarity 0.89)
- Open-Set Domain Adaptation for Semantic Segmentation (CVPR 2024 · similarity 0.88)
- COSDA: Counterfactual-based Susceptibility Risk Framework for Open-Set Domain Adaptation (ICML 2025 · similarity 0.87)
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.
