ClearPose
收藏资源简介:
ClearPose是由密歇根大学创建的大规模真实世界RGB-D透明物体数据集,旨在解决透明物体在视觉感知和深度估计中的挑战。该数据集包含超过350,000个标记的真实世界RGB-D帧和500万个实例注释,涵盖63种家用物品。数据集内容丰富,包括日常生活中的常见物品,以及多种光照和遮挡条件下的挑战性测试场景。创建过程中,利用了ProgressLabeller标注系统,确保了标注的效率和准确性。ClearPose数据集主要应用于透明物体的分割、场景级深度完成和以物体为中心的姿态估计任务,为相关领域的研究提供了重要的基准数据。
ClearPose is a large-scale real-world RGB-D transparent object dataset created by the University of Michigan, designed to address the challenges of transparent objects in visual perception and depth estimation. This dataset contains over 350,000 labeled real-world RGB-D frames and 5 million instance annotations, covering 63 types of household items. The dataset features diverse content, including common daily objects as well as challenging test scenarios under varying lighting and occlusion conditions. During its development, the ProgressLabeller annotation system was utilized to ensure the efficiency and accuracy of the annotations. The ClearPose dataset is primarily applied to tasks such as transparent object segmentation, scene-level depth completion, and object-centric pose estimation, providing critical benchmark data for research in relevant fields.

- 1ClearPose: Large-scale Transparent Object Dataset and Benchmark密歇根大学 · 2022年



