AQUA

问答 (QA) 对是使用最先进的问题生成方法自动生成的，该方法基于现有艺术理解数据集中提供的绘画和评论。 QA 对由众包工作人员根据其语法正确性、可回答性和答案的正确性进行清理。数据集本质上由视觉（基于绘画）和知识（基于评论）问题组成。Question-answer (QA) pairs were automatically generated using state-of-the-art question generation methods, which are based on the paintings and comments provided in existing art understanding datasets. The QA pairs were then cleaned by crowdworkers according to their grammatical correctness, answerability, and the correctness of the answers. The dataset essentially comprises visual (painting-based) and knowledge (comment-based) questions.

搜集汇总

背景与挑战

背景概述

AQUA数据集是一个公开的视觉问答数据集，基于艺术理解数据中的绘画和评论，通过先进方法自动生成问答对，并经过众包清理以确保质量。它包含视觉和知识问题，由卡内基梅隆大学等机构于2020年发布，适用于自然语言处理和问题生成任务。