CORE
收藏资源简介:
CORE是由武汉大学团队构建的全球首个百万级跨模态地理定位数据集,涵盖全球225个地理区域的1,034,786对跨视角图像-文本对。该数据集通过大视觉语言模型合成高质量场景描述,提供了前所未有的环境多样性和城市布局变化视角。数据采集覆盖全球各大陆的城乡环境,特别关注不同气候带和人文建筑风格的空间异质性。该数据集旨在解决复杂场景下的全球导航问题,为跨模态地理定位研究提供大规模基准。
CORE is the world's first million-scale cross-modal geolocation dataset developed by the research team at Wuhan University. It encompasses 1,034,786 pairs of cross-view image-text data spanning 225 geographic regions across the globe. This dataset generates high-quality scene descriptions using large vision-language models, providing unprecedented environmental diversity and perspectives on variations in urban layouts. The data collection covers both urban and rural environments across all continents, with special emphasis on the spatial heterogeneity of different climate zones and human architectural styles. This dataset is designed to tackle the challenge of global navigation in complex scenarios, serving as a large-scale benchmark for cross-modal geolocation research.
CORE 数据集概述
数据集基本信息
- 数据集名称:CORE
- 托管平台:GitHub
- 仓库地址:https://github.com/YtH0823/CORE
数据集描述
根据提供的README文件内容,该数据集详情页面未包含具体的描述信息。




