EmbodiedCity/BasicSpatialAbility

该数据集是为评估视觉语言模型（VLMs）的基本空间能力（BSAs）而创建的，基于心理测量学框架定义了五种基本空间能力：空间感知、空间关系、空间定向、心理旋转和空间可视化。数据集包含原始测试和正确答案，用于通过九项验证的心理测量实验对13种主流VLMs进行基准测试。研究发现VLMs在空间能力上存在显著差距，并提出了改进方法。数据集旨在为空间人工智能的发展提供评估基准和方法论视角。

This dataset is created to evaluate the Basic Spatial Abilities (BSAs) of Visual Language Models (VLMs), defining five BSAs based on a psychometric framework: Spatial Perception, Spatial Relation, Spatial Orientation, Mental Rotation, and Spatial Visualization. The dataset includes original tests and correct answers, used to benchmark 13 mainstream VLMs through nine validated psychometric experiments. The research reveals significant gaps in VLMs spatial abilities and proposes enhancement methods. The dataset aims to provide an evaluation benchmark and methodological perspective for the development of Spatial Artificial Intelligence.