Research map / Label-efficient and robust learning / Domain adaptation and generalization
Towards Effective Instance Discrimination Contrastive Loss for Unsupervised Domain Adaptation
Yixin Zhang, Zilei Wang, Junjie Li, Jiafan Zhuang, Zihan Lin
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 low-confidence samples only for IDCo loss instead of all samples (ablation) | GVB+FixMatch with IDCo · Office-Home UDA | accuracy: 73.0 -> 73.3 | Improved |
| use high-confidence samples for IDCo loss (ablation) | GVB+FixMatch with IDCo · Office-Home UDA | accuracy: 72.7 -> 72.3 | Hurt |
| add class relationship enhanced features (CRF) to low-confidence IDCo (ablation) | EIDCo · Office-Home UDA | accuracy: 73.3 -> 74.4 | Improved |
| add CRF to high-confidence samples (ablation) | IDCo loss · Office-Home UDA | accuracy: 72.3 -> 72.4 | No clear effect |
| add plain cross-domain mixup (ablation) | EIDCo · Office-Home UDA | accuracy: 74.4 -> 74.6 | Improved |
| use target-dominated mixup rather than plain mixup (ablation) | EIDCo · Office-Home UDA | accuracy: 74.6 -> 75.1 | Improved |
| use CRF similarity instead of original features for EIDCo (ablation) | EIDCo · Office-Home UDA | accuracy: 73.6 -> 75.1 | Improved |
| use CRF similarity instead of original features for EIDCo (ablation) | EIDCo · VisDA-2017 UDA | accuracy: 80.1 -> 82.0 | Improved |
| add EIDCo on top of FixMatch with strong ViT baseline | Baseline-B in SSRT + FixMatch · Office-Home UDA | accuracy: 83.07 -> 86.38 | Improved |
| add EIDCo alone instead of FixMatch to baseline | Baseline-B in SSRT · VisDA-2017 UDA | class-wise accuracy: FixMatch 86.61 vs EIDCo 86.37 | Hurt |
| add EIDCo to weaker baselines (Source only, DANN, CDAN) | weaker UDA baselines · Office-Home UDA | accuracy: 46.1 -> 52.6; 57.6 -> 60.3; 65.8 -> 67.6 | Improved |
| apply IDCo loss directly in feature space (even with projection head) | UDA baselines · UDA benchmarks | accuracy: < 0.5% | No clear effect |
Closest papers by meaning
- Mutual Nearest Neighbor Contrast and Hybrid Prototype Self-Training for Universal Domain Adaptation (AAAI 2022 · similarity 0.90)
- Instance Level Affinity-Based Transfer for Unsupervised Domain Adaptation (CVPR 2021 · similarity 0.90)
- One Fits Many: Class Confusion Loss for Versatile Domain Adaptation (TPAMI 2024 · similarity 0.89)
- Contrastive Adaptation Network for Unsupervised Domain Adaptation (CVPR 2019 · similarity 0.89)
- Guiding Pseudo-Labels With Uncertainty Estimation for Source-Free Unsupervised Domain Adaptation (CVPR 2023 · similarity 0.89)
- Low Category Uncertainty and High Training Potential Instance Learning for Unsupervised Domain Adaptation (AAAI 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.
