GRAB
收藏资源简介:
GRAB是由剑桥大学和香港大学联合创建的图表分析基准数据集,旨在评估大型多模态模型在图表分析任务中的性能。该数据集包含2170个合成生成的图表分析问题,覆盖四个核心任务和23个图表属性。数据集通过Matplotlib库合成生成,确保了高质量和无噪声的问题。GRAB主要应用于提升大型多模态模型在图表分析领域的性能,特别是在无法访问图表底层数据的情况下,通过视觉解释进行数据分析的能力。
GRAB is a chart analysis benchmark dataset jointly developed by the University of Cambridge and the University of Hong Kong, which aims to evaluate the performance of large multimodal models on chart analysis tasks. This dataset contains 2,170 synthetically generated chart analysis questions, covering four core tasks and 23 chart attributes. All samples are synthetically generated using the Matplotlib library, ensuring high-quality and noise-free question samples. GRAB is primarily utilized to enhance the performance of large multimodal models in the chart analysis domain, particularly their ability to conduct data analysis via visual interpretation without access to the underlying chart data.




