Research map / Vision-language and multimodal / CLIP and zero-shot
LoGoPrompt: Synthetic Text Images Can Be Good Visual Prompts for Vision-Language Models
Cheng Shi, Sibei Yang
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 |
|---|---|---|---|
| use synthetic class-name text images as visual prompts with min-max contrastive prompt selection, versus CoCoOp | LoGoPrompt · Base-to-new, 11 dataset average | harmonic mean: 75.83 -> 79.03 | Improved |
| use class-wise synthetic visual prompts rather than learned visual prompts (VPT, DPT) | CLIP base-to-new · Base-to-new, new-class accuracy | new-class accuracy: +3.88% | Improved |
| replace learned pixel-perturbation Visual Prompting with LoGoPrompt | CLIP RN50 few-shot · 16-shot, 11 dataset average | accuracy: 61.84 -> 76.31 | Improved |
| add feature adapter to LoGoPrompt | LoGoPrompt · 16-shot, 11 dataset average | accuracy: 76.31 -> 77.28 | Improved |
| apply LoGoPrompt instead of CoCoOp for domain generalization | CLIP ViT-B/16 · ImageNet to V2/Sketch/A/R | target average accuracy: 62.13 -> 63.82 | Improved |
| treat synthetic text images as extra training images (no prompt selection) (ablation) | LoGoPrompt · ImageNet 16-shot RN50 | accuracy: 63.12 | No clear effect |
| add visual prompt selection with min-max contrastive loss versus paired text-image (ablation) | LoGoPrompt · ImageNet 16-shot RN50 | accuracy: 63.54 -> 66.34 | Improved |
| use too large a visual prompt (ratio 2/7 vs 1/7) (ablation) | LoGoPrompt · ImageNet 16-shot RN50 | accuracy: 66.34 -> 66.17 | Hurt |
| use random location for visual prompt instead of fixed (ablation) | LoGoPrompt · ImageNet 16-shot RN50 | accuracy: 66.01 -> 66.34 | Improved |
| increase text prompt length M from 2 to 16 (ablation) | LoGoPrompt · ImageNet 16-shot RN50 | accuracy: 65.58 -> 66.34 | Improved |
Later papers that compared with this method
1 later accepted paper reports LoGoPrompt as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Causality-guided Prompt Learning for Vision-language Models via Visual Granulation (ICCV 2025 · better on 2 of 2 shared results)
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
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.90)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- Learning to Learn Better Visual Prompts (AAAI 2024 · similarity 0.89)
- Understanding prompt engineering may not require rethinking generalization (ICLR 2024 · similarity 0.89)
- LaViP: Language-Grounded Visual Prompting (AAAI 2024 · similarity 0.89)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (AAAI 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.
