工业技术文档比赛数据集

本次工业技术文档多模态推理问答评测任务，具有三大挑战：（1）图片型原始文档识别：不可编辑的PDF文档难以准确识别数据，分辨率低，格式多样，内容复杂。（2）多模态信息融合需求：问题解答常需同时解析文本描述和技术图纸的多模态数据；（3）复杂化领域知识推理：答案常需通过图纸结构解析、模块功能理解或机械原理推导获得。

This multimodal reasoning QA evaluation task for industrial technical documents presents three core challenges:
1. Image-based original document recognition: Non-editable PDF documents encounter difficulties in accurate data identification, featuring low resolution, diverse formats and complex content.
2. Multimodal information fusion requirement: Answering questions often requires simultaneous parsing of both textual descriptions and multimodal data from technical drawings.
3. Complex domain knowledge reasoning: Answers are usually derived through structural parsing of drawings, comprehension of module functions or deduction of mechanical principles.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于工业技术文档多模态推理问答评测的数据集，主要挑战包括图片型文档识别、多模态信息融合和复杂领域知识推理。数据集文件为IndustryDoc_VQA.zip，大小311.06MB。