Research map / Vision-language and multimodal / CLIP and zero-shot
PRO-VPT: Distribution-Adaptive Visual Prompt Tuning via Prompt Relocation
Chikai Shang, Mengke Li, Yiqun Zhang, Zhen Chen, Jinlin Wu, Fangqing Gu, Yang Lu, Yiu-Ming Cheung
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 |
|---|---|---|---|
| add adaptive prompt distribution via relocation (PRO-VPT) to VPT-Deep | VPT-Deep · VTAB-1k avg | accuracy: 76.4 -> 78.0 | Improved |
| add prompt relocation to VPT | VPT-Deep · FGVC avg | accuracy: 89.7 -> 91.7 | Improved |
| use relocation (G) instead of pruning matched to trained parameters (A) (ablation) | PRO-VPT · SVHN | accuracy: 83.76 -> 88.19 | Improved |
| use relocation (G) instead of pruning matched to trained params (A) (ablation) | PRO-VPT · Resisc45 | accuracy: 83.81 -> 86.56 | Improved |
| decompose relocation into pruning and allocation steps instead of naive RL (C) (ablation) | PRO-VPT · SVHN | accuracy: 83.51 -> 88.19 | Improved |
| decompose relocation into pruning and allocation steps instead of naive RL (C) (ablation) | PRO-VPT · Resisc45 | accuracy: 83.02 -> 86.56 | Improved |
| use idleness-score pruning (Prn2) instead of random pruning (Prn1) with PPO allocation (ablation) | PRO-VPT · SVHN | accuracy: 86.99 -> 88.19 | Improved |
| use idleness-score pruning (Prn2) instead of random pruning (Prn1) with PPO allocation (ablation) | PRO-VPT · Resisc45 | accuracy: 85.68 -> 86.56 | Improved |
| use PPO RL allocation (Alc2) instead of Thompson sampling bandit (Alc1) (ablation) | PRO-VPT · SVHN | accuracy: 85.82 -> 88.19 | Improved |
| use PPO RL allocation (Alc2) instead of Thompson sampling bandit (Alc1) (ablation) | PRO-VPT · Resisc45 | accuracy: 84.76 -> 86.56 | Improved |
| prune from huge hyper-distribution (B) instead of matched-parameter pruning (A) (ablation) | pruning baseline · SVHN | accuracy: 83.76 -> 84.17 | No clear effect |
| add PRO-VPT to ViT-L/16 backbone | VPT-Deep · VTAB-1k Cifar100 | accuracy: 85.82 -> 87.07 | Improved |
Closest papers by meaning
- Revisiting the Power of Prompt for Visual Tuning (ICML 2024 · similarity 0.92)
- Facing the Elephant in the Room: Visual Prompt Tuning or Full finetuning? (ICLR 2024 · similarity 0.91)
- DA-VPT: Semantic-Guided Visual Prompt Tuning for Vision Transformers (CVPR 2025 · similarity 0.90)
- LoR-VP: Low-Rank Visual Prompting for Efficient Vision Model Adaptation (ICLR 2025 · similarity 0.88)
- VIPAMIN: Visual Prompt Initialization via Embedding Selection and Subspace Expansion (NeurIPS 2025 · similarity 0.88)
- Enhancing Visual Prompting through Expanded Transformation Space and Overfitting Mitigation (NeurIPS 2025 · similarity 0.87)
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.
