Research map / Vision-language and multimodal / CLIP and zero-shot
uCAP: An Unsupervised Prompting Method for Vision-Language Models
A. Tuan Nguyen, Kai Sheng Tai, Bor-Chun Chen, Satya Narayan Shukla, Hanchao Yu, Philip Torr, Tai-Peng Tian, Ser-Nam Lim
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 |
|---|---|---|---|
| learn unsupervised domain prompts (10 prompts) instead of hand-tuned prompts | CLIP RN50 · SUN397 | accuracy: 59.5 -> 64.6 | Improved |
| replace pseudo-label prompt learning with energy-based uCAP | CLIP ViT-B/16 · ImageNet | accuracy: 66.1 -> 69.7 | Improved |
| use 10 learned prompts instead of 1 (ablation) | uCAP · ImageNet, CLIP RN50 | accuracy: 60.5 -> 61.9 | Improved |
| uCAP vs hand-tuned prompts on strong model | CLIPA ViT-H/14 · ImageNet | accuracy: 81.0 -> 81.0 | No clear effect |
| unsupervised uCAP vs labeled CoOp on ImageNetV2 | CLIP RN50 · ImageNetV2 | accuracy: 55.40 -> 54.70 | Hurt |
| unsupervised uCAP vs labeled CoOp on ImageNet-R | CLIP RN50 · ImageNet-R | accuracy: 56.60 -> 61.80 | Improved |
| use online learning instead of offline (ablation) | uCAP · ImageNet, CLIP RN50 | accuracy: 61.9 -> 61.4 | Hurt |
| transfer prompts learned on ImageNet to other tasks (ablation) | uCAP · SUN397, CLIP RN50 | accuracy: 53.1 -> 62.8 vs Classnames Only | Improved |
| add pseudo labels for prompt learning | CLIP · ImageNet, RN50 | accuracy: 54.8 -> 57.6 | Improved |
Closest papers by meaning
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024 · similarity 0.92)
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.90)
- Domain-Agnostic Mutual Prompting for Unsupervised Domain Adaptation (CVPR 2024 · similarity 0.90)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (AAAI 2024 · similarity 0.90)
- ProAPO: Progressively Automatic Prompt Optimization for Visual Classification (CVPR 2025 · similarity 0.89)
- UMFC: Unsupervised Multi-Domain Feature Calibration for Vision-Language Models (NeurIPS 2024 · 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.
