Chart Instruction-Tuning Dataset

该数据集包含了122,857张多样化的图表图像，用于生成视觉指令调整数据，以支持图表理解和推理任务。该数据集特别关注视觉属性和指令质量，在反映图表内容的准确性上达到了82%。覆盖的规模达到了122,857张图表图像，涉及的任务包括思维链、摘要、事实核查、图表到Markdown转换以及程序辅助设计等多种指令调整任务。

This dataset comprises 122,857 diverse chart images, which are employed to generate visual instruction tuning data to support chart understanding and reasoning tasks. It places special focus on visual attributes and instruction quality, achieving an 82% accuracy rate in reflecting the content of the charts. With a total of 122,857 chart images, the dataset covers a variety of instruction tuning tasks including chain-of-thought, summarization, fact-checking, chart-to-Markdown conversion, and program-aided design.