AI Paper Atlas

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

ICCV 2025 · Vision-language and multimodal › CLIP and zero-shot · approach: Parameter-efficient visual tuning · 4 citations (OpenAlex)

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 changedWhereMeasuredEffect
add adaptive prompt distribution via relocation (PRO-VPT) to VPT-DeepVPT-Deep · VTAB-1k avgaccuracy: 76.4 -> 78.0Improved
add prompt relocation to VPTVPT-Deep · FGVC avgaccuracy: 89.7 -> 91.7Improved
use relocation (G) instead of pruning matched to trained parameters (A) (ablation)PRO-VPT · SVHNaccuracy: 83.76 -> 88.19Improved
use relocation (G) instead of pruning matched to trained params (A) (ablation)PRO-VPT · Resisc45accuracy: 83.81 -> 86.56Improved
decompose relocation into pruning and allocation steps instead of naive RL (C) (ablation)PRO-VPT · SVHNaccuracy: 83.51 -> 88.19Improved
decompose relocation into pruning and allocation steps instead of naive RL (C) (ablation)PRO-VPT · Resisc45accuracy: 83.02 -> 86.56Improved
use idleness-score pruning (Prn2) instead of random pruning (Prn1) with PPO allocation (ablation)PRO-VPT · SVHNaccuracy: 86.99 -> 88.19Improved
use idleness-score pruning (Prn2) instead of random pruning (Prn1) with PPO allocation (ablation)PRO-VPT · Resisc45accuracy: 85.68 -> 86.56Improved
use PPO RL allocation (Alc2) instead of Thompson sampling bandit (Alc1) (ablation)PRO-VPT · SVHNaccuracy: 85.82 -> 88.19Improved
use PPO RL allocation (Alc2) instead of Thompson sampling bandit (Alc1) (ablation)PRO-VPT · Resisc45accuracy: 84.76 -> 86.56Improved
prune from huge hyper-distribution (B) instead of matched-parameter pruning (A) (ablation)pruning baseline · SVHNaccuracy: 83.76 -> 84.17No clear effect
add PRO-VPT to ViT-L/16 backboneVPT-Deep · VTAB-1k Cifar100accuracy: 85.82 -> 87.07Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy