This summary presents GUIRepair Hu25, a cross-modal reasoning approach for automated program repair that addresses visual software issues through bidirectional transformation between textual and visual modalities. Originally published at ASE 2025, this work introduces the first systematic solution for resolving multimodal GitHub issues involving graphical user interfaces, achieving state-of-the-art performance on the SWE-bench M benchmark with 157 resolved instances using GPT-4o and 175 instances with o4-mini.
Kai Huang (Thu,) studied this question.