When MLLMs miss what humans read: diagnosing and learning to read squint text

作者：Chen Peng, Shan Huang, Jianbo Huang, Linrui Qin, Qibo Sun, Mengwei Xu · 单位：State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing, China · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-10-12