SprayD
收藏资源简介:
SprayD是由弗莱堡大学团队开发的专注于非朗伯体物体深度补全的基准数据集,包含398帧多传感器同步采集的RGB-D数据。该数据集通过工业级哑光喷雾技术和三摄像头(Intel RealSense D415、ZED2、Azure Kinect DK)融合方案,首次实现了场景级稠密真实深度标注,覆盖非朗伯体物体及周边环境。数据采集采用机器人臂控制标准化轨迹,结合后期多视角深度融合与人工校验,最终获得97.8%稠密度的精准标注。该数据集旨在解决透明/反光物体深度感知难题,为机器人抓取、导航等任务提供可靠的几何基准。
SprayD is a benchmark dataset developed by the team at the University of Freiburg, focusing on depth completion for non-Lambertian objects. It contains 398 frames of multi-sensor synchronously acquired RGB-D data. This dataset adopts industrial-grade matte spray technology and a fusion scheme with three cameras (Intel RealSense D415, ZED2, Azure Kinect DK), and for the first time achieves scene-level dense real-world depth annotations covering non-Lambertian objects and their surrounding environments. Data collection was conducted using a robotic arm to control standardized trajectories, combined with post-processing multi-view depth fusion and manual verification, ultimately producing accurate annotations with a density of 97.8%. This dataset aims to solve the depth perception challenges of transparent or reflective objects, providing reliable geometric benchmarks for tasks such as robotic grasping and navigation.




