Research map / Label-efficient and robust learning / Domain adaptation and generalization
Measure The Feature Universe: Topology-based Pseudo Labeling and Gravity Consistency for Source-Free Domain Adaptation
Jae Yun Lee, Hyeok Nam, Sung In Cho
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 universe-aware pseudo labeling to SHOT (ablation) | SHOT · Office-Home (average) | accuracy: 71.9 -> 72.5 | Improved |
| add gravity consistency to SHOT (ablation) | SHOT · Office-Home (average) | accuracy: 71.9 -> 73.0 | Improved |
| combine universe-aware pseudo labeling with gravity consistency (ablation) | SHOT · Office-Home (average) | accuracy: 71.9 -> 74.4 | Improved |
| add gravity consistency without information maximization loss (ablation) | weighted CE · DomainNet-126 (average) | accuracy: 68.6 -> 66.3 | Hurt |
| add gravity consistency together with information maximization loss (ablation) | CE + IM · DomainNet-126 (average) | accuracy: 68.9 -> 73.6 | Improved |
| add virtual features (lambda 0 to 20) (ablation) | universe-aware pseudo labeling · Office-Home (average) | accuracy: 73.8 -> 74.4 | Improved |
| increase virtual features from 20 to 30 (ablation) | universe-aware pseudo labeling · Office-Home (average) | accuracy: 74.4 -> 74.0 | Hurt |
| use sqrt schedule for quantile instead of fixed values (ablation) | reliable area scheduling · DomainNet-126 (average) | accuracy: 73.1 -> 73.6 | Improved |
| use sqrt schedule instead of linear schedule (ablation) | reliable area scheduling · DomainNet-126 (average) | accuracy: 73.3 -> 73.6 | Improved |
| add KL-based consistency regularization to SHOT | SHOT · Office-Home A to P | accuracy: 78.8 -> 80.2 | Improved |
| replace KL consistency with gravity consistency | SHOT · Office-Home A to P | accuracy: 80.2 -> 80.9 | Improved |
| add gravity consistency to existing SFDA methods | SHOT, GKD, TPDS · DomainNet-126 (average) | accuracy: 68.1 -> 73.2 (SHOT) | Improved |
Closest papers by meaning
- Unraveling the Mysteries of Label Noise in Source-Free Domain Adaptation: Theory and Practice (TPAMI 2025 · similarity 0.89)
- BMD: A General Class-Balanced Multicentric Dynamic Prototype Strategy for Source-Free Domain Adaptation (ECCV 2022 · similarity 0.88)
- Source-Free Domain Adaptation Guided by Vision and Vision-Language Pre-training (IJCV 2024 · similarity 0.88)
- Source-Free Progressive Graph Learning for Open-Set Domain Adaptation (TPAMI 2023 · similarity 0.88)
- Pseudo-Labeling for Unsupervised Domain Adaptation with Kernel GLMs (NeurIPS 2026 · similarity 0.86)
- HiGDA: Hierarchical Graph of Nodes to Learn Local-to-Global Topology for Semi-Supervised Domain Adaptation (AAAI 2025 · similarity 0.86)
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.
