Calibrate Before Reason: Robust Visual Token Reduction against Semantic Drift in VLMs 文章

ArXiv CS.CV2026-07-31PAPERen作者: Jiasheng Li, Zhong Ji, Yan Zhang, Huihui Li
大语言模型
查看原文 →
关系图谱

Calibrate Before Reason: Robust Visual Token Reduction against Semantic Drift in VLMs · 相关公司

A
ANDINONPROFIT
A
AMI团队RESEARCH_INSTITUTE