egenioussBench
收藏资源简介:
egenioussBench是由奥地利技术研究院与布伦瑞克工业大学联合开发的城市场景视觉定位基准数据集,包含高分辨率航空3D网格和CityGML LoD2模型两种地理空间参考数据。数据集核心为2,709张智能手机拍摄的厘米级精度地面图像,通过专业摄影测量束调整技术获取独立于地图的真实坐标,并从中提取412张序列化验证图像和42张非共视测试图像。其创新性在于首次将航空网格与简化城市模型结合,支持跨视角、跨域定位算法验证,旨在解决传统SfM方法在大规模场景中存储成本高、扩展性差的核心问题,为自动驾驶、无人机导航等应用提供标准化评估平台。
egenioussBench is a visual localization benchmark dataset for urban scenes developed jointly by the Austrian Institute of Technology and Technische Universität Braunschweig. It includes two types of geospatial reference data: high-resolution aerial 3D meshes and CityGML LoD2 models. The core of the dataset comprises 2,709 ground-level images captured by smartphones with centimeter-level accuracy. The ground truth coordinates independent of external maps are acquired through professional photogrammetric bundle adjustment, and 412 sequential validation images and 42 non-co-visible test images are extracted from this collection. Its innovation lies in the first integration of aerial 3D meshes and CityGML LoD2 models, which supports the verification of cross-view and cross-domain localization algorithms. This dataset aims to address the core issues of traditional Structure from Motion (SfM) methods in large-scale scenes, namely high storage costs and poor scalability, and provides a standardized evaluation platform for applications such as autonomous driving and unmanned aerial vehicle (UAV) navigation.
根据您提供的页面HTML文本内容,该页面是一个名为“egeniouss”的欧盟项目主页,并非一个具体的数据集详情页面。因此,无法从中提炼出任何数据集的相关信息。
页面内容主要介绍该项目本身,其核心是一个实时视觉定位云服务平台,旨在提升卫星导航的精度、可靠性和平台独立性,但未提及任何可下载或使用的特定数据集(如训练数据、基准数据集等)。页面中提到的“机器学习”、“地理数据”、“众包”等技术组件和“egenioussBench”链接,均未提供具体数据集的描述、规模、格式或访问方式。
鉴于上述情况,无法根据该HTML内容生成数据集详情总结。




