RoboPoint Synthetic Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://robo-point.github.io
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个在模拟环境中生成的逼真图像集，包含了大约66万张图像与关系配对，这些配对来自于1万个程序生成的随机场景，并使用可供性点进行了标注。此外，该数据集在自由空间中包含了可供性标签，旨在为机器人任务提升数据多样性。其规模达到了66万图像-关系配对，任务是对空间可供性进行预测。
This dataset is a collection of photorealistic images generated in a simulated environment, containing approximately 660,000 image-relation pairs. These pairs are derived from 10,000 programmatically generated random scenes and annotated with affordance points. Additionally, the dataset includes affordance labels in free space, with the objective of enhancing data diversity for robotic tasks. It has a total scale of 660,000 image-relation pairs, and the targeted task is spatial affordance prediction.
