microsoft/VISION_LANGUAGE

该数据集是一个合成数据集，用于测试模型在空间推理、导航和计数方面的能力。它包含三种任务：空间理解、导航和计数。每种任务都有三种输入模态：仅文本、仅视觉和视觉+文本。该数据集中的问题类型有三种：判断一个对象相对于另一个对象的方向、识别某个方向上的对象以及计算某个方向上的对象数量，所有问题均为选择题。

This dataset is a synthetic dataset designed to test models abilities in spatial reasoning, navigation, and counting. It includes three tasks: Spatial Understanding, Navigation, and Counting. Each task has three input modalities: text-only, vision-only, and vision-text. The dataset features three types of questions: determining the direction of one object from another, identifying an object in a certain direction, and counting the number of objects in a direction, with all questions being multiple-choice.