Research map / Vision-language and multimodal / CLIP and zero-shot
Efficient Test-Time Prompt Tuning for Vision-Language Models
Yuhan Zhu, Guozhen Zhang, Xu Chen, Haocheng Shen, Xiaoxin Chen, Gangshan Wu, Limin Wang
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 CPT during source prompt learning (ablation) | CoOp · Base-to-new (Specialized) | accuracy: 67.6 -> 70.5 | Improved |
| add test-time adaptation with CPT (ablation) | CoOp+CPT · Base-to-new (Specialized) | accuracy: 70.5 -> 73.3 | Improved |
| add gradient matching loss (ablation) | CoOp+CPT+TTA · Base-to-new (Specialized) | accuracy: 73.3 -> 74.9 | Improved |
| add Front view to CPT views (ablation) | CPT · Base-to-new (Specialized) | accuracy: 74.4 -> 74.9 | Improved |
| add Mid view to CPT views (ablation) | CPT · Base-to-new (Specialized) | accuracy: 73.9 -> 74.9 | Improved |
| add End view to CPT views (ablation) | CPT · Base-to-new (Specialized) | accuracy: 73.7 -> 74.9 | Improved |
| add hand-made prompt view to CPT (ablation) | CPT · Base-to-new (Specialized) | accuracy: 73.6 -> 74.9 | Improved |
| use cosine instead of MSE gradient matching (ablation) | GM loss · Base-to-new (Specialized) | accuracy: 73.9 -> 74.9 | Improved |
| add EMA to gradient matching (ablation) | GM loss · Base-to-new (Specialized) | accuracy: 73.4 -> 74.9 | Improved |
| combine MaPLe with TPT | MaPLe · Cross-dataset generalization | accuracy: 66.30 -> 66.50 | No clear effect |
| combine MaPLe with TPT | MaPLe · Base-to-new generalization | accuracy: 75.14 -> 74.95 | Hurt |
| combine CoOp with TPT | CoOp · Base-to-new generalization | accuracy: 73.06 -> 73.77 | Improved |
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.
- MaPLe — MaPLe: Multi-Modal Prompt Learning (CVPR 2023) · 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
- Black-Box Test-Time Prompt Tuning for Vision-Language Models (AAAI 2025 · similarity 0.91)
- Self-Prompt Mechanism for Few-Shot Image Recognition (AAAI 2024 · similarity 0.90)
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.89)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- On the Test-Time Zero-Shot Generalization of Vision-Language Models: Do We Really Need Prompt Learning? (CVPR 2024 · similarity 0.89)
- ZIP: An Efficient Zeroth-order Prompt Tuning for Black-box Vision-Language Models (ICLR 2025 · similarity 0.89)
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.
