GraspNet-1Billion is an RGB-D image dataset for object grasping posture detection, containing 190 complex backgrounds and 97,280 images, each with precise 6D posture annotations and object grasping posture annotations, a total of 88 objects and more than 1.1 billion grasping postures. These images are taken by two mainstream RGB-D cameras, Kinect Azure and RealSense D435.

This dataset can be used to study general object grasping and other related areas such as 6D pose estimation, unseen object segmentation, etc.