Research map / Label-efficient and robust learning / Test-time adaptation
Dynamic Logits Adjustment and Exploration for Test-Time Adaptation in Vision Language Models
Haoyan Wu, Yahao Liu, Yinjie Lei, Lixin Duan, 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 Dynamic Logits Adjustment to DPE-style baseline (ablation) | DLAE (ViT-B/16) · Cross-dataset average | top-1 accuracy: 69.40 -> 70.63 | Improved |
| add CGEC on top of DLA (ablation) | DLAE (ViT-B/16) · Cross-dataset average | top-1 accuracy: 70.63 -> 71.88 | Improved |
| add entropy adjustment in DLA training (ablation) | DLAE (ViT-B/16) · Cross-dataset average | top-1 accuracy: 69.40 -> 69.85 | Improved |
| add logits adjustment at inference (ablation) | DLAE (ViT-B/16) · Cross-dataset average | top-1 accuracy: 69.40 -> 70.39 | Improved |
| combine entropy and logits adjustment (ablation) | DLAE (ViT-B/16) · Cross-dataset average | top-1 accuracy: 70.39 -> 70.63 | Improved |
| add Semantic Consistency Filter to CGEC (ablation) | DLAE (ViT-B/16) · Cross-dataset average | top-1 accuracy: 70.63 -> 71.26 | Improved |
| add Temporal Consistency Filter to CGEC (ablation) | DLAE (ViT-B/16) · Cross-dataset average | top-1 accuracy: 70.63 -> 71.37 | Improved |
| add both SCF and TCF (ablation) | DLAE (ViT-B/16) · Cross-dataset average | top-1 accuracy: 70.63 -> 71.88 | Improved |
| vary alpha and beta (ablation) | DLAE (ViT-B/16) · DTD | top-1 accuracy: 56.58 to 58.86 | No clear effect |
| vary eta from 0.002 to 0.05 (ablation) | DLAE (ViT-B/16) · DTD, Aircraft, Cars | top-1 accuracy | No clear effect |
| use DLAE instead of DPE | DLAE (ViT-B/16) · ImageNet-A (ViT-B/16) | top-1 accuracy: 59.63 -> 64.06 | Improved |
| use DLAE instead of SCA on ImageNet-R | DLAE (ViT-B/16) · ImageNet-R (ViT-B/16) | top-1 accuracy: 80.85 -> 80.43 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BoostAdapter — BoostAdapter: Improving Vision-Language Test-Time Adaptation via Regional Bootstrapping (NeurIPS 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Efficient Test-Time Adaptation of Vision-Language Models (CVPR 2024 · similarity 0.94)
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.93)
- DOTA: Distributional Test-time Adaptation of Vision-Language Models (NeurIPS 2025 · similarity 0.93)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.92)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.92)
- Realistic Test-Time Adaptation of Vision-Language Models (CVPR 2025 · similarity 0.91)
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.
