Research map / Vision-language and multimodal / CLIP and zero-shot
Learning to Prompt for Vision-Language Models
Kaiyang Zhou, Jingkang Yang, Chen Change Loy, Ziwei Liu
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 |
|---|---|---|---|
| replace hand-crafted prompt with learned context (CoOp, M=16) | Zero-Shot CLIP ResNet-50 · ImageNet | accuracy: 58.18 -> 62.95 | Improved |
| use fewer context tokens (M=4 vs 16) for domain shift (ablation) | CoOp ResNet-50 · ImageNet-Sketch | accuracy: 32.74 -> 34.67 | Improved |
| apply CoOp (M=4) to ImageNet-A transfer | CLIP ResNet-50 · ImageNet-A | accuracy: 21.65 -> 23.06 | Improved |
| replace zero-shot with linear probe | CLIP ResNet-50 · ImageNet-Sketch | accuracy: 33.32 -> 19.07 | Hurt |
| use learned CoOp prompts instead of prompt ensembling | CLIP ResNet-50 · ImageNet | accuracy: 60.41 -> 62.95 | Improved |
| use prompt ensembling instead of single engineered prompt | Zero-Shot CLIP ViT-B/16 · ImageNet | accuracy: 66.73 -> 68.74 | Improved |
| fine-tune CLIP image encoder instead of prompt learning | CLIP ResNet-50 · ImageNet (16 shots) | accuracy: 58.18 -> 18.28 | Hurt |
| add learnable transformation layer on text output | Zero-Shot CLIP ResNet-50 · ImageNet (16 shots) | accuracy: 58.18 -> 58.86 | Improved |
| add learnable bias on text output | Zero-Shot CLIP ResNet-50 · ImageNet (16 shots) | accuracy: 58.18 -> 60.93 | Improved |
| manual vs random initialization of context (ablation) | CoOp · Average over 11 datasets | accuracy: 72.65 -> 72.65 | No clear effect |
| add CoOp learned context to CLIP ResNet-50 on DOSCO-2k | CLIP ResNet-50 · DOSCO-2k | accuracy: 60.9 -> 64.7 | Improved |
| use CoCoOp instead of CoOp on DOSCO-2k | CoOp ViT-B/16 · DOSCO-2k | accuracy: 73.3 -> 72.7 | Hurt |
Closest papers by meaning
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.93)
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.92)
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024 · similarity 0.92)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (AAAI 2024 · similarity 0.91)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- Weighted Multi-Prompt Learning with Description-free Large Language Model Distillation (ICLR 2025 · similarity 0.90)
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.
