Research map / Learning theory and algorithms / Optimal transport
Visual Prompting via Partial Optimal Transport
Mengyu Zheng, Zhiwei Hao, Yehui Tang, Chang Xu
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 frequency-based label mapping with optimal-transport label mapping | visual prompting ViT-B · 10 downstream datasets average | accuracy: 61.47 -> 69.77 | Improved |
| replace frequency-based label mapping with OTLM on ResNet50 | visual prompting ResNet50 · 10 downstream datasets average | accuracy: 34.77 -> 37.48 | Improved |
| use OT cost matrix instead of frequency cost with greedy solver (ablation) | OTLM · Flowers102 | accuracy: 62.56 -> 79.92 | Improved |
| use linear programming solver instead of greedy with OT cost (ablation) | OTLM · Flowers102 | accuracy: 79.92 -> 86.63 | Improved |
| use linear programming instead of greedy with frequency cost (ablation) | FLM/ILM style mapping · Flowers102 | accuracy: 62.56 -> 64.29 | Improved |
| apply OTLM under VPT without extra head versus FLM/ILM | Swin-B VPT · 5 datasets average | accuracy: 81.6 -> 84.30 | Improved |
| apply OTLM versus VP on small datasets | ViT-B VP · datasets ordered by samples per class | performance gain over VP | Improved |
| use larger ImageNet-22K source model instead of ImageNet-1K for label mapping (ablation) | OTLM, FLM, ILM · CUB200 | performance gain over VP: average cost OTLM 13.53 -> 34.18; FLM 17.96 -> 43.85 | Mixed |
| OTLM on SVHN and Food101 versus ILM-VP | OTLM · SVHN (ViT-B), Food101 (ResNet50) | accuracy: SVHN 88.8 -> 88.7; Food101 22.2 -> 21.8 | No clear effect |
Closest papers by meaning
- Bayesian-guided Label Mapping for Visual Reprogramming (NeurIPS 2024 · similarity 0.87)
- AutoVP: An Automated Visual Prompting Framework and Benchmark (ICLR 2024 · similarity 0.86)
- Exploring Task-Level Optimal Prompts for Visual In-Context Learning (AAAI 2025 · similarity 0.86)
- Query Efficient Black-Box Visual Prompting with Subspace Learning (CVPR 2025 · similarity 0.86)
- SA²VP: Spatially Aligned-and-Adapted Visual Prompt (AAAI 2024 · similarity 0.86)
- Enhancing Visual Prompting through Expanded Transformation Space and Overfitting Mitigation (NeurIPS 2025 · similarity 0.86)
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.
