Research map / Vision-language and multimodal / CLIP and zero-shot
Multi-Cache Enhanced Prototype Learning for Test-Time Generalization of Vision-Language Models
Xinyu Chen, Haotian Zhai, Can Zhang, Xiupeng Shi, Ruirui 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 vision-text prototype residual learning with align and contrast losses | MCP · Cross-domain 10 datasets average (ViT-B/16) | top-1 accuracy (%): 71.98 -> 72.61 | Improved |
| add residual learning (MCP++ over MCP) | MCP · ImageNet (ViT-B/16) | top-1 accuracy (%): 72.37 -> 72.64 | Improved |
| add residual learning (MCP++ over MCP) | MCP · Cross-domain 10 datasets average (ResNet-50) | top-1 accuracy (%): 65.08 -> 65.49 | Improved |
| add negative cache to entropy and align caches (ablation) | MCP++ · ImageNet (ViT-B/16) | top-1 accuracy (%): 72.49 -> 72.64 | Improved |
| add align cache to entropy and negative caches (ablation) | MCP++ · ImageNet (ViT-B/16) | top-1 accuracy (%): 71.73 -> 72.64 | Improved |
| add entropy cache to align and negative caches (ablation) | MCP++ · ImageNet (ViT-B/16) | top-1 accuracy (%): 72.54 -> 72.64 | Improved |
| add any single cache over CLIP baseline (ablation) | MCP++ · ImageNet (ViT-B/16) | top-1 accuracy (%): 66.73 -> 71.51 | Improved |
| add alignment loss (ablation) | MCP++ · Aircraft | top-1 accuracy (%): 29.74 -> 30.07 | Improved |
| add contrastive loss (ablation) | MCP++ · Aircraft | top-1 accuracy (%): 29.74 -> 29.92 | Improved |
| add both align and contrast losses (ablation) | MCP++ · Aircraft | top-1 accuracy (%): 29.74 -> 30.18 | Improved |
| use multi-retrieval strategy over single retrieval (ablation) | MCP++ · ImageNet | top-1 accuracy (%): +3.16% | Improved |
| use multi-retrieval strategy over single retrieval (ablation) | MCP++ · Cross-domain 10 datasets | top-1 accuracy (%): +4.24% | 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
- Dual Prototype Evolving for Test-Time Generalization of Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- Efficient Test-Time Adaptation of Vision-Language Models (CVPR 2024 · similarity 0.88)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.88)
- BoostAdapter: Improving Vision-Language Test-Time Adaptation via Regional Bootstrapping (NeurIPS 2024 · similarity 0.88)
- DOTA: Distributional Test-time Adaptation of Vision-Language Models (NeurIPS 2025 · similarity 0.88)
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.
