1 paper
Kaixiang Zhao, Tianrun Yu, Shawn Huang +3
We study fact-level repair for multimodal generation, where a fluent output may contain specific facts that are not supported by the input. Existing inference-time repair methods o…