Plot2Code
收藏资源简介:
Plot2Code是由腾讯ARC实验室创建的一个综合视觉编码基准数据集,旨在评估多模态大型语言模型(MLLMs)在从科学图表生成代码方面的能力。该数据集包含132个精心挑选的高质量matplotlib图表,涵盖六种图表类型,每个图表都配有其源代码和由GPT-4总结的描述性指令。Plot2Code不仅评估MLLMs的代码生成能力,还提出了三种自动评估指标,包括代码通过率、文本匹配比率和GPT4V整体评分,以细致评估输出代码和渲染图像。数据集的应用领域主要集中在评估和提升MLLMs在视觉编码任务中的表现,特别是在处理文本密集型图表时的能力。
Plot2Code is a comprehensive visual coding benchmark dataset created by Tencent ARC Lab, which aims to evaluate the code generation capabilities of multimodal large language models (MLLMs) from scientific charts. This dataset contains 132 high-quality, carefully curated matplotlib charts spanning six chart types. Each chart is accompanied by its corresponding source code and descriptive instructions summarized by GPT-4. In addition to evaluating the code generation performance of MLLMs, Plot2Code also introduces three automatic evaluation metrics, namely code pass rate, text matching ratio, and GPT4V overall score, to conduct a fine-grained assessment of the output code and its rendered images. The primary application scenarios of this dataset focus on evaluating and enhancing the performance of MLLMs in visual coding tasks, particularly their ability to handle text-intensive charts.




