Research map / Vision-language and multimodal / CLIP and zero-shot
CARPRT: Class-Aware Zero-Shot Prompt Reweighting for Vision-Language Model
Ruijiang Dong, Zesheng Ye, Jianzhong Qi, Lei Feng, Feng Liu, Gang Niu, Masashi Sugiyama
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 class-specific prompt weights (CARPRT) instead of class-agnostic WPE | CLIP-ViT-B/16 prompt ensemble · Zero-shot average (11 datasets) | accuracy (%): 65.13 -> 67.45 | Improved |
| use class-specific weights instead of class-averaged global weights (CARPRT-Uniform) (ablation) | CARPRT · 10 fine-grained datasets | accuracy: +2.39% average gain | Improved |
| apply CARPRT instead of WPE on specialized domains | CLIP-ViT-B/16 · Aircraft | accuracy (%): 23.28 -> 24.49 | Improved |
| transfer CARPRT weights from ImageNet to variants | CLIP-ViT-B/16 · ImageNet-A/R/Sketch/V2 average | accuracy (%): 61.08 -> 61.73 | Improved |
| add iterative refinement of pseudo-labels and weights (iCARPRT) (ablation) | CARPRT · 10 fine-grained datasets, ViT-B/16 | accuracy (%): 67.34 -> 67.48 | Mixed |
| add confidence- or entropy-based pseudo-label filtering (ablation) | CARPRT · zero-shot classification | accuracy | No clear effect |
| lower temperature tau below 1.0 (ablation) | CARPRT · zero-shot classification | accuracy | Hurt |
| use CARPRT as prompt pool selector instead of human selection with TDA | TDA · 10 datasets, ViT-B/16 | accuracy (%): 67.53 -> 68.02 | Improved |
| use CARPRT instead of human selection with TDA | TDA · 10 datasets, ResNet50 | accuracy (%): 63.82 -> 63.76 | No clear effect |
Closest papers by meaning
- Improved Zero-Shot Classification by Adapting VLMs with Text Descriptions (CVPR 2024 · similarity 0.89)
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024 · similarity 0.89)
- Visual and Semantic Prompt Collaboration for Generalized Zero-Shot Learning (CVPR 2025 · similarity 0.89)
- ProAPO: Progressively Automatic Prompt Optimization for Visual Classification (CVPR 2025 · similarity 0.88)
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.88)
- Bridge the Modality and Capability Gaps in Vision-Language Model Selection (NeurIPS 2024 · 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.
