Our work on failure prediction of vision-language models has been accepted at ICCV 2025 :star: !