Leodocr/egovista-pilot-v0.2
收藏资源简介:
EgoVista Pilot v0.2是一个关于人类日常生活操作活动的第一人称(自我中心)注释数据集。该数据集仅包含注释,不包含RGB视频帧。注释内容涵盖每帧的2D姿势关键点(全身)、2D和3D手部关键点、16位深度图、手部和物体分割掩码、接触标志(左/右/双手)以及动作标签(动词+对象,由Gemini 2.5 Pro生成)。数据集包含11个片段,共6,448帧,约10.7分钟的源素材,帧率为10 fps。数据格式为LeRobot v3.0(Parquet + 元数据),源分辨率为1920x1080。数据集主要用于机器人学和视频分类任务,语言为英语,标签包括自我中心、操作等。数据集大小为1K到10K之间,许可证为CC-BY-NC 4.0,仅限非商业使用。
EgoVista Pilot v0.2 is a first-person (egocentric) annotations dataset of human daily-life manipulation activities, generated by the EgoVista pipeline. This is an annotations-only release, with no RGB video frames included. Per-frame annotations cover 2D pose keypoints (full body), 2D and 3D hand keypoints, 16-bit depth maps, hand and object segmentation masks, contact flags (left/right/both hands), and action labels (verb + object, generated by Gemini 2.5 Pro). The dataset includes 11 episodes, 6,448 frames, approximately 10.7 minutes of source footage at 10 fps. The format is LeRobot v3.0 (Parquet + metadata), with a source resolution of 1920x1080. It is intended for tasks in robotics and video classification, in English, with tags such as egocentric, manipulation, etc. The size category is 1K<n<10K, licensed under CC-BY-NC 4.0 for non-commercial use only.




