MTabVQA
收藏arXiv2025-06-13 更新2025-11-28 收录
下载链接：
https://hf-mirror.com/datasets/mtabvqa/MTabVQA-Eval
下载链接
链接失效反馈官方服务：
资源简介：
MTabVQA是一个专门设计用于评估多表格视觉问答能力的基准数据集。该数据集包含3745个复杂的问题-答案对，需要跨越多个视觉渲染的表格图像进行多跳推理。数据集的创建过程包括数据收集、数据提取和预处理、视觉表格渲染、多跳问答对生成以及验证和筛选。MTabVQA旨在解决视觉语言模型在处理现实世界场景中多表格数据时的推理能力问题。
MTabVQA is a benchmark dataset specifically designed to evaluate multi-table visual question answering capabilities. This dataset contains 3,745 complex question-answer pairs that require multi-hop reasoning across multiple visually rendered table images. The creation of the MTabVQA dataset includes data collection, data extraction and preprocessing, visual table rendering, generation of multi-hop question-answer pairs, as well as validation and filtering. MTabVQA aims to address the reasoning capability issues of vision-language models when handling multi-table data in real-world scenarios.
提供机构：
1Department of Information Technology, Panjab University, India 2Language Technology Group, Universität Hamburg, Germany
创建时间：
2025-06-13
