Research map / Vision-language and multimodal / CLIP and zero-shot
PLOT: Prompt Learning with Optimal Transport for Vision-Language Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 multiple prompts with OT feature alignment rather than a single CoOp prompt | CLIP RN50 · Few-shot average, 1 shot | accuracy: 59.56±2.06 -> 62.59±1.13 | Helped |
| learn multiple prompts with OT feature alignment rather than a single CoOp prompt | CLIP RN50 · Few-shot average, 16 shots | accuracy: 73.33±0.42 -> 73.94±0.54 | Helped |
| match multiple prompts to the global visual feature without OT (ablation) | PLOT · Caltech101, 1 shot | accuracy: 89.83±0.33 -> 88.13±0.36 | Hurt |
| add prompt-distance variance regularization (ablation) | global-feature multiple-prompt method G · Caltech101, 1 shot | accuracy: 88.13±0.36 -> 88.28±0.43 | No clear effect |
| use diverse predefined prompt initializations (ablation) | global-feature multiple-prompt method G · Caltech101, 1 shot | accuracy: 88.13±0.36 -> 88.91±0.38 | Helped |
| replace global features with feature maps and remove OT (ablation) | PLOT · Caltech101, 1 shot | accuracy: 89.83±0.33 -> 69.78±1.75 | Hurt |
| increase the prompt count from 1 to 4 (ablation) | PLOT · Caltech101, 1 shot | accuracy: 88.47±1.15 -> 89.83±0.33 | Helped |
| increase the prompt count from 4 to 8 (ablation) | PLOT · Food101, 1 shot | accuracy: 77.74±0.47 -> 78.05±0.15 | Helped |
| apply PLOT to Tip-Adapter-F | Tip-Adapter-F · Few-shot average, 16 shots | accuracy: 75.83 -> 76.20 | Helped |
| apply PLOT prompts and OT to CoCoOp | CoCoOp · Base-to-new average, new classes | accuracy: 64.6 -> 67.6 | Helped |
| use OT with engineered prompts instead of global-feature cosine matching | CLIP zero-shot · ImageNet | top-1 accuracy: 60.38% -> 58.78% | Hurt |
Closest papers by meaning
- Learning to Prompt with Text Only Supervision for Vision-Language Models (AAAI 2025 · similarity 0.90)
- ProAPO: Progressively Automatic Prompt Optimization for Visual Classification (CVPR 2025 · similarity 0.90)
- PromptKD: Unsupervised Prompt Distillation for Vision-Language Models (CVPR 2024 · similarity 0.90)
- IPO: Interpretable Prompt Optimization for Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- CoPL: Contextual Prompt Learning for Vision-Language Understanding (AAAI 2024 · similarity 0.90)
- NLPrompt: Noise-Label Prompt Learning for Vision-Language Models (CVPR 2025 · 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.