FigureQA

FigureQA是一个专为视觉推理研究设计的大型数据集，包含超过100,000张合成科学风格图像和超过100万的问题-答案对。数据集涵盖五种常见的图表类型：线图、点线图、垂直和水平条形图、饼图。数据集通过15种模板生成问题，涉及图表元素之间的关系和特征，如最大值、最小值、曲线下的面积、平滑度和交点。数据集还包括用于机器学习系统训练的辅助数据，如生成每个图表的数值数据和所有图表元素的边界框注释。FigureQA旨在解决机器从视觉表示中直观识别模式的问题，适用于机器学习和人工智能领域的研究。

FigureQA is a large-scale dataset specifically developed for visual reasoning research, containing over 100,000 synthetic scientific-style images and more than 1 million question-answer pairs. The dataset covers five common chart types: line charts, line-scatter plots, vertical and horizontal bar charts, and pie charts. Questions are generated based on 15 templates, involving relationships and features between chart elements such as maximum value, minimum value, area under the curve, smoothness, and intersection points. The dataset also includes auxiliary data for training machine learning systems, including the numerical data used to generate each chart and bounding box annotations for all chart elements. FigureQA aims to address the challenge of enabling machines to intuitively recognize patterns from visual representations, and is applicable to research in the fields of machine learning and artificial intelligence.