MultiChartQA
收藏资源简介:
MultiChartQA 是一个用于评估多模态大语言模型(MLLMs)在多图表场景中能力的基准数据集。该数据集由西安交通大学和圣母大学的研究团队创建,包含来自ArXiv、OECD、OWID和Pew Research Center的655个图表和944个问题。数据集设计了四种问题类型,涵盖直接问答、并行问答、比较推理和顺序推理,旨在全面评估模型在多图表理解中的能力。数据集的创建过程包括从多个公共资源中收集图表,并手动注释问题和答案,以确保高质量。MultiChartQA 主要应用于多图表理解和视觉问答领域,旨在解决现有基准在复杂多图表场景中的不足。
MultiChartQA is a benchmark dataset for evaluating the capabilities of multimodal large language models (MLLMs) in multi-chart scenarios. Developed by a research team from Xi'an Jiaotong University and the University of Notre Dame, it includes 655 charts and 944 questions sourced from ArXiv, OECD, OWID, and Pew Research Center. The dataset features four question types, covering direct question answering, parallel question answering, comparative reasoning, and sequential reasoning, aiming to comprehensively assess models' multi-chart understanding capabilities. The development process of the dataset entails collecting charts from multiple public resources and manually annotating both questions and answers to ensure high quality. Primarily applied in the fields of multi-chart understanding and visual question answering, MultiChartQA aims to address the limitations of existing benchmarks in complex multi-chart scenarios.
Multi-chart QA 数据集概述
简介
Multi-chart QA 是一个广泛且具有挑战性的基准测试数据集,包含真实世界的图表。数据集中的图表来自多个来源,以确保多样性和完整性。每个多图表组包含2到3个图表,每个组伴随4个问题,通常涵盖四个不同的主要类别。
数据集规模
- 数据集包含236个多图表组。
- 总计包含944个问题。

- 1MultiChartQA: Benchmarking Vision-Language Models on Multi-Chart Problems西安交通大学, 圣母大学 · 2024年



