Vision-language models (VLMs) are increasingly used as automated judges for multimodal and embodied tasks, yet aggregate accuracy or consistency alone does n…
机构:中科大
来源:arXiv 2611.13596 | AI4Papers 论文推荐平台