Appear2Meaning
收藏资源简介:
Appear2Meaning是由曼彻斯特大学、武汉大学及盖蒂保护研究所联合构建的跨文化多类别文化遗产基准数据集,包含750件来自盖蒂博物馆和大都会艺术博物馆的文物图像及结构化元数据(如创作者、时期、地理起源等)。数据集覆盖东亚、古代地中海、欧洲和美洲四大文化区域,涵盖陶瓷、绘画、金属制品和雕塑四类文物,每类样本经严格人工验证。其旨在评估视觉语言模型从图像推断非直观文化属性的能力,解决文化遗产领域结构化元数据自动标注的难题,推动跨文化多模态理解研究。
Appear2Meaning is a cross-cultural, multi-category cultural heritage benchmark dataset jointly developed by the University of Manchester, Wuhan University, and the Getty Conservation Institute. It contains 750 cultural heritage images and structured metadata (including creator, period, geographic origin and other relevant information) sourced from the Getty Museum and the Metropolitan Museum of Art. The dataset covers four cultural regions: East Asia, Ancient Mediterranean, Europe and the Americas, and encompasses four categories of cultural relics: ceramics, paintings, metalware and sculptures, with each sample undergoing strict manual verification. It aims to evaluate the capability of vision-language models to infer non-intuitive cultural attributes from images, address the challenges of automatic annotation of structured metadata in the cultural heritage domain, and promote research on cross-cultural multimodal understanding.




