GraspNet

GraspNet 是一个不断丰富的通用对象抓取的开放项目。它目前包含：

GraspNet-1Billion：包含 190 个由两个流行的 RGBD 相机（Kinect Azure 和 RealSense D435）捕获的杂乱和复杂场景的基准测试，总共带来 97,280 张图像。对于每张图像，我们标注准确的 6D 姿势和每个对象的密集抓取姿势。总的来说，我们的数据集包含 88 个对象和超过 11 亿个抓取姿势。我们希望我们的基准能够促进一般对象抓取以及其他相关领域（例如 6D 姿势估计、看不见的对象分割等）的相关研究。

GraspNet is an ever-growing open project dedicated to general-purpose object grasping. Currently, it includes:
GraspNet-1Billion: a benchmark consisting of 190 cluttered and complex scenes captured by two popular RGBD cameras, namely Kinect Azure and RealSense D435, with a total of 97,280 images. For each image, we have annotated accurate 6D poses and dense grasping poses for every object. Overall, this dataset encompasses 88 object categories and over 1.1 billion grasping poses. We hope that this benchmark will promote relevant research in general object grasping and other related fields, such as 6D pose estimation, unseen object segmentation, and so on.

搜集汇总

背景与挑战

背景概述

GraspNet是一个专注于通用对象抓取的大规模数据集，包含GraspNet-1Billion基准测试，提供了190个杂乱场景的97,280张RGBD图像，标注了88个对象的6D姿势和超过11亿个抓取姿势。该数据集旨在促进对象抓取、6D姿势估计和看不见的对象分割等领域的研究，由上海交通大学于2020年发布，采用CC BY-NC-SA 4.0许可证。