VisText
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/microsoft/lmm-graphical-perception
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为VisText，涵盖了包括体育和新闻在内的多个领域，其中既包含文本数据表格，也包含用于生成各种图表的示例Vega-Lite程序。数据集包含了名义、序数和数值属性，每个数据集最多有两个维度和5个数据点。规模上，该数据集包含了1,000个随机抽取的数据集。任务是对使用生成的图表来评估和诊断机器学习模型的图形感知能力。
This dataset is named VisText, spanning multiple domains such as sports and news. It encompasses both tabular text data and sample Vega-Lite programs for generating diverse types of charts. The constituent datasets feature nominal, ordinal, and numerical attributes, with each containing at most two dimensions and 5 data points. Overall, VisText includes 1,000 randomly sampled datasets. The task of this dataset is to evaluate and diagnose the graphical perception capabilities of machine learning models via the generated charts.
提供机构：
Microsoft
