SuctionNet-1Billion

SuctionNet-1Billion是由上海交通大学创建的大规模真实世界吸盘抓取数据集，包含超过10亿个吸盘抓取标注。该数据集通过两步法生成，首先对88个单个物体进行吸盘姿态采样和标注，然后利用物体的6D姿态生成190个杂乱场景的密集标注。每个场景包含512张RGB-D图像，总计97,280张图像。数据集旨在为吸盘抓取算法提供大规模、高密度且与真实机器人试验对齐的标注，以推动吸盘抓取技术的发展。

SuctionNet-1Billion is a large-scale real-world suction cup grasping dataset created by Shanghai Jiao Tong University, containing over 1 billion annotated suction grasping instances. This dataset is generated via a two-stage pipeline: first, suction pose sampling and annotation are conducted for 88 individual objects; then, dense annotations for 190 cluttered scenes are generated using the 6D poses of the objects. Each scene contains 512 RGB-D images, totaling 97,280 images across all scenes. The dataset aims to provide large-scale, high-density annotations aligned with real robotic trials for suction cup grasping algorithms, so as to promote the development of suction grasping technology.