Foveated Probes Recover Localized Binding Information in Vision Foundation Models

作者：Mateusz Michalkiewicz, Mahsa Baktashmotlagh, Guha Balakrishnan · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-08-04