FinChart-Bench
收藏资源简介:
FinChart-Bench是一个专注于现实世界金融图表的大型视觉语言模型理解基准数据集。该数据集包含从2015年到2024年收集的1200个金融图表图像,每个图像都标注了True/False、Multiple Choice和Question Answering三种类型的问题,共计7016个问题。FinChart-Bench旨在解决当前大型视觉语言模型在金融图表理解任务中的能力不足问题。数据集的创建经历了两个阶段的严格人工评估,以确保数据的质量和准确性。FinChart-Bench的数据集适用于评估和改进大型视觉语言模型在金融图表理解任务中的性能。
FinChart-Bench is a large-scale visual-language model understanding benchmark dataset focused on real-world financial charts. It contains 1,200 financial chart images collected from 2015 to 2024, with each image annotated with three types of questions: True/False, Multiple Choice, and Question Answering, totaling 7,016 questions. FinChart-Bench aims to address the insufficient capabilities of current large-scale visual-language models in financial chart understanding tasks. The dataset was created through two stages of rigorous manual evaluation to ensure its quality and accuracy. The FinChart-Bench dataset is suitable for evaluating and improving the performance of large-scale visual-language models on financial chart understanding tasks.
FinChart-Bench 数据集概述
数据集基本信息
- 任务类别: 图像文本到文本 (image-text-to-text)
- 标签: 金融 (financial)、图表 (charts)、基准测试 (benchmark)、视觉语言模型 (vision-language-models)、问答 (question-answering)
数据集内容
- 数据量: 1,200 张金融图表图像
- 时间范围: 2015 年至 2024 年
- 标注类型:
- 真/假问题 (True/False, TF)
- 多项选择题 (Multiple Choice, MC)
- 问答题 (Question Answering, QA)
- 问题总数: 7,016 个
相关资源




