StereOBJ-1M
收藏资源简介:
StereOBJ-1M数据集是由卡内基梅隆大学创建的大型立体RGB图像对象姿态估计数据集,旨在解决透明、半透明和反射物体的姿态估计挑战。该数据集包含超过393,000帧和超过150万个6D对象姿态注释,涵盖18个对象在182个场景中的记录。数据集创建过程中,采用了一种新颖的多视角方法来高效标注姿态数据,使得数据捕获可以在复杂和灵活的环境中进行。该数据集主要应用于增强现实和机器人操作等领域,旨在解决透明和反射物体的姿态估计问题。
StereOBJ-1M is a large-scale stereo RGB image object pose estimation dataset created by Carnegie Mellon University, aiming to address the challenges of pose estimation for transparent, translucent and reflective objects. This dataset contains over 393,000 frames and more than 1.5 million 6D object pose annotations, covering 18 objects across 182 recording scenes. During the dataset construction, a novel multi-view method was adopted to efficiently annotate pose data, enabling data capture to be conducted in complex and flexible environments. This dataset is primarily applied in fields such as augmented reality and robotic manipulation, targeting the pose estimation problem of transparent and reflective objects.




