Berkeley MVP Data
收藏资源简介:
Berkeley MVP Data 是由加州大学伯克利分校(University of California, Berkeley)创建的机器人学习数据集,专注于真实世界机器人任务的视觉预训练。该数据集包含来自互联网和第一人称视频的大量图像,以及在真实环境中收集的机器人操作数据。数据集的创建过程涉及从多个数据源收集图像,包括ImageNet、Epic Kitchens、Something Something、100 Days of Hands和Ego4D等数据集,总计450万张图像。这些数据用于通过掩码自动编码器(MAE)进行自监督视觉预训练,以学习有用的视觉表示。
Berkeley MVP Data is a robotics learning dataset developed by the University of California, Berkeley, focusing on visual pre-training for real-world robotic tasks. This dataset includes a vast collection of images sourced from the internet and first-person videos, alongside robotic manipulation data gathered in real-world environments. The dataset creation process involves collecting images from multiple data sources, including datasets such as ImageNet, Epic Kitchens, Something-Something, 100 Days of Hands, and Ego4D, with a total of 4.5 million images. These data are used for self-supervised visual pre-training via Masked Autoencoders (MAE) to learn useful visual representations.




