Research map / Label-efficient and robust learning / Test-time adaptation
Probe, Anchor, and Amend: Active Test-Time Adaptation of Vision-Language Models
Jiaqi Lin, Chaoqi Chen, Xiasi Wang, Mingfu Yan, Jiancheng Huang, Jianzhuang Liu, Wenming Yang, Qingmin Liao
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 alignment loss L_align (ablation) | PAA · Natural Distribution Shifts avg, ViT-B/16 | top-1 accuracy (%): 67.80 -> 68.19 | Improved |
| add intra-class compactness loss L_intra (ablation) | PAA · Natural Distribution Shifts avg, ViT-B/16 | top-1 accuracy (%): 67.92 -> 68.19 | Improved |
| add inter-class separation loss L_inter (ablation) | PAA · Natural Distribution Shifts avg, ViT-B/16 | top-1 accuracy (%): 68.03 -> 68.19 | Improved |
| add memory-driven amend loss L_amend (ablation) | PAA · Natural Distribution Shifts avg, ViT-B/16 | top-1 accuracy (%): 67.98 -> 68.19 | Improved |
| train visual LayerNorm parameters instead of freezing (ablation) | PAA · Natural Distribution Shifts avg, ViT-B/16 | top-1 accuracy (%): 67.28 -> 68.19 | Improved |
| train textual residuals instead of freezing (ablation) | PAA · Natural Distribution Shifts avg, ViT-B/16 | top-1 accuracy (%): 67.38 -> 68.19 | Improved |
| use textual residuals instead of text-side LayerNorm (ablation) | PAA · Natural Distribution Shifts avg, ViT-B/16 | top-1 accuracy (%): 67.61 -> 68.19 | Improved |
| increase prototype cache size from 1 to 3 (ablation) | PAA · ImageNet-A, ViT-B/16 | top-1 accuracy (%): 65.22 -> 66.59 | Improved |
| increase prototype cache size from 3 to 7 (ablation) | PAA · ImageNet-A, ViT-B/16 | top-1 accuracy (%): 66.59 -> 65.52 | Hurt |
| increase memory lifetime from 1 to 3 (ablation) | PAA · ImageNet-A, ViT-B/16 | top-1 accuracy (%): 66.32 -> 66.59 | Improved |
| increase memory lifetime from 3 to 5 (ablation) | PAA · ImageNet-A, ViT-B/16 | top-1 accuracy (%): 66.59 -> 66.63 | No clear effect |
| use full-class labels instead of binary verification (ablation) | PAA · Natural Distribution Shifts avg, ViT-B/16 | top-1 accuracy (%): 68.19 -> 68.84 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- BoostAdapter — BoostAdapter: Improving Vision-Language Test-Time Adaptation via Regional Bootstrapping (NeurIPS 2024) · better on 4 of 4 shared results
- BATCLIP — BATCLIP: Bimodal Online Test-Time Adaptation for CLIP (ICCV 2025) · better on 1 of 1 shared result
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
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.93)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.93)
- DOTA: Distributional Test-time Adaptation of Vision-Language Models (NeurIPS 2025 · similarity 0.92)
- Dual Prototype Evolving for Test-Time Generalization of Vision-Language Models (NeurIPS 2024 · similarity 0.92)
- Efficient Test-Time Adaptation of Vision-Language Models (CVPR 2024 · similarity 0.92)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.92)
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.
