Raffael-Kultyshev/hand_tracking_lerobot_dataset

这是一个专业级的机器人手部追踪数据集，采用LeRobot格式，专为训练AI驱动的机器人系统设计，适用于Figure AI和Physical Intelligence等公司。数据集包含从视频记录中提取的高质量手部追踪数据，每个片段包括：第一人称视角的手部操作任务视频、8个关键手部关节的3D坐标和方向角度、时间同步的关节位置数据以及完整的片段信息和数据集统计。数据集统计包括总片段数、总帧数、帧率、关节数和状态向量大小等。数据集结构包括Parquet文件存储的片段数据、视频文件、元数据和README文件。每个关节的维度包括3D位置和方向角度。数据格式详细说明了Parquet文件中的列和状态向量布局。使用示例包括通过LeRobot、Pandas和HuggingFace Datasets加载数据集，以及使用自定义可视化工具和LeRobot可视化工具进行数据可视化。数据集收集使用了iPhone摄像头和MediaPipe手部追踪技术，处理过程包括手部标志点提取、3D位置归一化和方向角度计算。质量说明提到某些帧可能包含NaN值，数据与视频逐帧同步，关节位置归一化，方向角度以度为单位。训练分割包括训练集和测试集。使用要求包括Python 3.7+和相关库的安装。

Professional-grade hand tracking dataset in LeRobot format, designed for training AI-driven robotics systems for companies like Figure AI and Physical Intelligence. This dataset contains high-quality hand tracking data extracted from video recordings using MediaPipe. Each episode includes: first-person perspective recordings showing hand manipulation tasks, 3D coordinates (x, y, z) and orientation angles (roll, pitch, yaw) for 8 key hand joints, time-synchronized joint positions across all frames, and complete episode information and dataset statistics. Dataset statistics include total episodes, total frames, FPS, number of joints, and state vector size. The dataset structure includes Parquet files for episode data, video files, metadata, and README file. Each joints dimensions include 3D position and orientation angles. The data format details the columns in Parquet files and the state vector layout. Usage examples include loading the dataset via LeRobot, Pandas, and HuggingFace Datasets, and visualizing the data using custom visualizer and LeRobot visualizer tools. Data collection used iPhone cameras and MediaPipe hand tracking, with processing including hand landmark extraction, 3D position normalization, and orientation angle calculation. Quality notes mention possible NaN values in some frames, frame-by-frame synchronization with video, normalized joint positions, and orientation angles in degrees. Training splits include train and test sets. Requirements include Python 3.7+ and installation of relevant libraries.