Research map / Label-efficient and robust learning / Test-time adaptation
Contrastive Test-Time Adaptation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 AdaContrast instead of SHOT | test-time adaptation · DomainNet-126, 7-shift Avg. | accuracy: 67.1 -> 67.8 | Helped |
| use AdaContrast instead of SHOT | test-time adaptation · VisDA-C train → val | per-class average accuracy: 83.0 -> 86.8% | Helped |
| add pseudo labeling to the source-only model (ablation) | AdaContrast · DomainNet-126, 7-shift Avg. | accuracy: 55.6 -> 58.5 | Helped |
| add pseudo labeling to the source-only model (ablation) | AdaContrast · VisDA-C train → val | per-class average accuracy: 43.8 -> 55.0 | Helped |
| add online pseudo-label refinement (ablation) | pseudo-labeling adaptation · DomainNet-126, 7-shift Avg. | accuracy: 58.5 -> 64.7 | Helped |
| add online pseudo-label refinement (ablation) | pseudo-labeling adaptation · VisDA-C train → val, learning rate 1x | per-class average accuracy: 55.0 -> 86.5 | Helped |
| add online pseudo-label refinement (ablation) | pseudo-labeling adaptation · VisDA-C train → val, learning rate 10x | per-class average accuracy: 44.5 -> 9.9 | Hurt |
| add joint contrastive learning (ablation) | online-refined pseudo-label adaptation · DomainNet-126, 7-shift Avg. | accuracy: 64.7 -> 67.9 | Helped |
| add joint contrastive learning (ablation) | online-refined pseudo-label adaptation · VisDA-C train → val, learning rate 10x | per-class average accuracy: 9.9 -> 84.3 | Helped |
| add weak-strong consistency and diversity regularization (ablation) | AdaContrast with joint contrastive learning · DomainNet-126, 7-shift Avg. | accuracy: 67.9 -> 67.8 | No clear effect |
| add weak-strong consistency and diversity regularization (ablation) | AdaContrast with joint contrastive learning · VisDA-C train → val, learning rate 10x | per-class average accuracy: 84.3 -> 86.6 | Helped |
| exclude same-class negative pairs using pseudo labels (ablation) | AdaContrast contrastive learning · DomainNet-126 and VisDA-C | accuracy: +0.2%, +2.1%, and +2.8% | Helped |
Closest papers by meaning
- Improved Self-Training for Test-Time Adaptation (CVPR 2024 · similarity 0.87)
- Domain-Specific Block Selection and Paired-View Pseudo-Labeling for Online Test-Time Adaptation (CVPR 2024 · similarity 0.87)
- Test-Time Domain Adaptation by Learning Domain-Aware Batch Normalization (AAAI 2024 · similarity 0.87)
- Efficient Test-Time Adaptation of Vision-Language Models (CVPR 2024 · similarity 0.87)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.86)
- A Layer Selection Approach to Test Time 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.