MMKE-Bench
收藏资源简介:
MMKE-Bench是一个全面的多模态知识编辑基准,由山东大学等机构的研究人员创建,旨在评估大型多模态模型在现实世界场景中编辑多样化视觉知识的能力。该数据集包含2940条知识和8363幅图像,涵盖了33个广泛类别,通过自动生成和人工验证的评价问题进行评估。数据集整合了视觉实体编辑、视觉语义编辑和用户特定编辑三种类型的任务,使用自由形式的自然语言来表示和编辑知识,为多模态知识编辑技术的评估设置了新的标准。
MMKE-Bench is a comprehensive multimodal knowledge editing benchmark developed by researchers from Shandong University and other institutions. It is designed to evaluate the capability of large multimodal models to edit diverse visual knowledge in real-world scenarios. The dataset comprises 2940 knowledge entries and 8363 images, spanning 33 broad categories, and is evaluated using automatically generated and manually verified evaluation questions. Three types of tasks are integrated into the dataset: visual entity editing, visual semantic editing, and user-specific editing. By utilizing free-form natural language to represent and edit knowledge, MMKE-Bench sets a new standard for the evaluation of multimodal knowledge editing technologies.




