Research map / Label-efficient and robust learning / Domain adaptation and generalization
Open Set Domain Adaptation by Backpropagation
Kuniaki Saito, Shohei Yamamoto, Yoshitaka Ushiku, Tatsuya Harada
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 adversarial unknown-class boundary (t=0.5) instead of OSVM on source-trained features | open set domain adaptation · Office 10 classes (AlexNet) | OS accuracy (%): 40.6 -> 80.4 | Improved |
| use adversarial unknown-class boundary instead of OSVM | open set domain adaptation · Office 10 classes (VGGNet) | OS accuracy (%): 65.3 -> 88.0 | Improved |
| add MMD distribution matching before OSVM | OSVM · Office 10 classes (AlexNet) | OS accuracy (%): 40.6 -> 34.5 | Hurt |
| add domain-classifier (BP) distribution matching before OSVM | OSVM · Office 10 classes (AlexNet) | OS accuracy (%): 40.6 -> 29.5 | Hurt |
| add MMD distribution matching before OSVM | OSVM · Office 10 classes (VGGNet) | OS accuracy (%): 65.3 -> 66.3 | No clear effect |
| use adversarial unknown-class boundary instead of OSVM | open set domain adaptation · Office 20 classes (VGGNet) | OS accuracy (%): 61.8 -> 74.7 | Improved |
| use adversarial unknown-class boundary instead of OSVM | open set domain adaptation · VisDA (AlexNet) | Avg accuracy over classes (%): 37.8 -> 58.5 | Improved |
| use adversarial unknown-class boundary instead of OSVM | open set domain adaptation · VisDA (VGGNet) | Avg accuracy over classes (%): 52.5 -> 65.2 | Improved |
| use adversarial unknown-class boundary instead of BP + OSVM | open set domain adaptation · Digits | UNK accuracy (%): 15.3 -> 85.9 | Improved |
| use adversarial unknown-class boundary instead of OSVM | open set domain adaptation · Digits | OS accuracy (%): 59.1 -> 82.4 | Improved |
| increase unknown-class target probability t toward 1 (ablation) | adversarial open set DA · Office DSLR to Amazon | OS and OS* accuracy: decrease as t increases | Hurt |
| add OSVM on top of the learned features | Ours · Semi-supervised open set recognition (SVHN) | UNK accuracy (%): 19.2 -> 93.1 | Improved |
Closest papers by meaning
- Separate to Adapt: Open Set Domain Adaptation via Progressive Separation (CVPR 2019 · similarity 0.91)
- Open Set Domain Adaptation (ICCV 2017 · similarity 0.91)
- Universal Source-Free Domain Adaptation (CVPR 2020 · similarity 0.90)
- Unknown-Aware Domain Adversarial Learning for Open-Set Domain Adaptation (NeurIPS 2022 · similarity 0.90)
- Weakly Supervised Open-Set Domain Adaptation by Dual-Domain Collaboration (CVPR 2019 · similarity 0.90)
- Distinguish Then Exploit: Source-free Open Set Domain Adaptation via Weight Barcode Estimation and Sparse Label Assignment (CVPR 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.
