MMT-Bench
收藏arXiv2024-04-25 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2404.16006v1
下载链接
链接失效反馈官方服务：
资源简介：
MMT-Bench是一个综合基准，旨在通过大量需要专业知识和精细视觉识别、定位、推理及规划的多模态任务来评估大型视觉-语言模型。该数据集包含31,325个精心策划的多选视觉问题，来自车辆驾驶和实体导航等多种多模态场景，涵盖32个核心元任务和162个子任务。
MMT-Bench is a comprehensive benchmark designed to evaluate large vision-language models via a diverse set of multimodal tasks that require professional expertise, fine-grained visual recognition, localization, reasoning and planning. This dataset contains 31,325 carefully curated multiple-choice visual questions, sourced from various multimodal scenarios such as vehicle driving and entity navigation, covering 32 core meta-tasks and 162 sub-tasks.
创建时间：
2024-04-25
