Slot2Text: Object-Centric Visual Tokenization for Efficient and Spatially Traceable Surgical MLLMs

作者：Guiqiu Liao, Matjaz Jogan, Daniel A. Hashimoto · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-08-04