AQUA (Art QUestion Answering)
收藏资源简介:
AQUA数据集是由大阪大学和卡内基梅隆大学合作创建的,专注于艺术作品的视觉问答任务。该数据集包含79,848条问答对,这些对是通过先进的问答生成方法自动从艺术作品及其评论中生成的。数据集分为视觉和知识两类问题,旨在测试模型对艺术作品视觉内容和背景知识的理解能力。创建过程中,使用了多种问答生成技术,并通过对众包工人的评估来确保问答对的质量。AQUA数据集的应用领域包括艺术理解、视觉识别和自然语言处理,旨在解决艺术领域中的视觉问答问题。
The AQUA dataset was co-created by Osaka University and Carnegie Mellon University, focusing on the visual question answering (VQA) task for artworks. It contains 79,848 question-answer pairs, which are automatically generated from artworks and their accompanying art critiques using advanced question answering generation methods. The dataset is categorized into two types of questions: visual and knowledge-based, aiming to evaluate models' ability to comprehend both the visual content and background knowledge of artworks. During its development, multiple question answering generation techniques were employed, and the quality of the question-answer pairs was ensured through evaluations conducted by crowd workers. The AQUA dataset has applications in fields including art understanding, visual recognition and natural language processing, with the core goal of addressing visual question answering problems in the art domain.




