MANBench/MANBench
收藏Hugging Face2025-02-11 更新2025-02-15 收录
下载链接：
https://hf-mirror.com/datasets/MANBench/MANBench
下载链接
链接失效反馈官方服务：
资源简介：
MANBench（多模态能力规范基准）是一个旨在评估人类和MLLMs（多模态语言模型）多模态能力的大型综合基准。该数据集包含9个任务，每个任务超过110个问题，共有1314个问题和2231张图片。MANBench的目标是提供一个公平和严格的评估框架，确保人类和机器性能的比较是在平等的基础上进行的。
MANBench (Multimodal Ability Norms Benchmark) is a comprehensive benchmark designed to evaluate the multimodal capabilities of both humans and MLLMs. The dataset consists of 9 tasks, each containing more than 110 questions, with a total of 1,314 questions and 2,231 images. MANBench aims to provide a fair and rigorous assessment framework, ensuring that comparisons between human and machine performance are conducted on an equitable basis.
提供机构：
MANBench
