manus-egocentric-sample
收藏资源简介:
该数据集包含以自我为中心(第一人称视角)的人类手部执行各种操作任务的记录,使用了Manus Metagloves进行高精度手指跟踪(约70Hz),OAK-D相机捕获RGB视频(1920x1080,30fps)和深度数据(640x400,30fps),以及IMU(加速度计和陀螺仪)数据。数据集统计信息包括总集数(20集)、总帧数(96,544帧)、帧率(30fps)、机器人类型(manus_gloves)、LeRobot版本(v2.1)和数据集大小(约15.3 GB)。数据集支持的任务包括'Pick_and_pack_task'和'fold_laundry'。数据集的特征包括集索引、帧索引、时间戳、任务索引、以自我为中心的图像、手部关节状态、手指角度、手势和IMU数据。此外,数据集还包含原始深度数据,格式为原始uint16二进制,分辨率为[400, 640](H, W),单位为毫米。数据集的使用方法包括通过LeRobot加载数据集,并提供了数据结构的详细描述和手部骨架结构的说明。
This dataset contains egocentric (first-person perspective) recordings of human hands performing various manipulation tasks. High-precision finger tracking (≈70 Hz) is conducted using Manus Metagloves, while an OAK-D camera captures RGB video (1920x1080, 30 fps), depth data (640x400, 30 fps), and IMU (accelerometer and gyroscope) data. The dataset statistics include total number of episodes (20 episodes), total frame count (96,544 frames), frame rate (30 fps), robot type (manus_gloves), LeRobot version (v2.1), and total dataset size (≈15.3 GB). The supported tasks for this dataset are 'Pick_and_pack_task' and 'fold_laundry'. The dataset features include episode index, frame index, timestamp, task index, egocentric images, hand joint states, finger angles, gestures, and IMU data. Additionally, the dataset includes raw depth data in uint16 binary format, with a resolution of [400, 640] (H, W) and units in millimeters. The usage instructions for this dataset include loading the dataset via LeRobot, with detailed descriptions of the data structure and explanations of the hand skeleton structure provided.



