When Attention Collapses: Stage-Aware Visual Token Pruning from Structure to Semantics 文章

ArXiv CS.CV2026-06-03NEWSen作者: Jiahui Wang, Kai Zhang, Mai Han, Huanghe Zhang
查看原文 →
关系图谱

When Attention Collapses: Stage-Aware Visual Token Pruning from Structure to Semantics · 相关技术

相关技术

Visual token pruningStructure-to-SemanticsVision-Language Models