AI RESEARCH
When Attention Collapses: Stage-Aware Visual Token Pruning from Structure to Semantics
arXiv CS.CV
•
ArXi:2606.03569v1 Announce Type: new Vision-Language Models (VLMs) have nstrated remarkable capabilities but suffer from significant computational overhead during inference. While visual token pruning offers a promising solution, existing methods predominantly rely on initial attention scores. This single-metric paradigm presents a critical flaw: high attention scores inherently collapse onto semantically similar regions, thereby severely reducing feature diversity and discarding vital contextual details. To address this, we.