GeoText-1652
收藏资源简介:
GeoText-1652是一个专为自然语言引导的无人机地理定位设计的基准数据集,由新加坡国立大学Sea-NExT联合实验室开发。该数据集通过结合大型语言模型和预训练视觉模型的交互式人机过程构建,扩展了现有的University-1652图像数据集,增加了空间感知文本标注,建立了图像、文本和边界框元素之间的一对一对应关系。数据集主要用于无人机通过文本进行导航和无人机视角目标定位,旨在解决现有数据集在语言引导无人机导航方面的不足,提供更精确的空间关系匹配,以提升无人机在现实世界场景中的控制和导航能力。
GeoText-1652 is a benchmark dataset specifically designed for natural language-guided drone geolocation, developed by the Sea-NExT Joint Lab at the National University of Singapore. This dataset is constructed via an interactive human-machine workflow that integrates large language models and pre-trained visual models. It expands the existing University-1652 image dataset by adding spatially-aware textual annotations, and establishes one-to-one correspondence among image, text, and bounding box elements. The dataset is mainly applied to text-based drone navigation and drone-view target localization, aiming to address the shortcomings of existing datasets in language-guided drone navigation, providing more precise spatial relationship matching to enhance the control and navigation capabilities of drones in real-world scenarios.




