MagmaAI/Magma-Video-ToM

Magma是一个用于多模态AI代理的基础模型。该数据集包含了用于Magma预训练的机器人操作数据，遵循OpenVLA的数据混合方式。数据集由多个源数据集组成，每个源包含一个或多个arrow文件。数据集字段包括数据集名称、视频名称、任务描述、视频片段的起始和结束时间戳、帧索引、图像的高度和宽度、机器人轨迹迹线以及迹线的可见性掩码。

Magma is a foundation model for multimodal AI agents. This dataset contains robotic manipulation data used in Magma pretraining, following the data mix of OpenVLA. The dataset is composed of multiple source datasets, each containing one or more arrow files. The dataset fields include dataset name, video name, task description, start and end timestamps of the video segment, frame index, image height and width, robot trajectory trace, and trace visibility mask.