MyDeadlift Dataset: Labeled Deadlift Videos and Biomechanical Angle Data

The MyDeadlift dataset contains 191 labeled videos of deadlift repetitions performed by seven male participants with different anthropometric characteristics and resistance-training experience. The recordings comprise three movement classes: Correct Movement (Gerakan Benar; GB; 83 videos), Rounded Back (Punggung Bungkuk; PB; 54 videos), and Knees Beyond Toes (Lutut Lebih Jari Kaki; LLK; 54 videos). Each video represents one complete repetition and was recorded from a fixed side-view position using the rear camera of an Apple iPhone 13 at 1920 × 1080-pixel resolution and 30 frames per second. The dataset is divided using a subject-wise splitting protocol. The training subset contains 131 videos from five participants, while the testing subset contains 60 videos from two participants, with 20 testing videos for each movement class. The repository also provides a processed table containing 21,813 frame-level records. Each record includes the data split, movement label, source video, frame number, knee angle, hip angle, and back-to-vertical angle. These biomechanical angles were calculated from body keypoints extracted using MediaPipe Pose. The dataset can be reused for human pose estimation, video-based exercise assessment, biomechanical movement analysis, human activity recognition, and the development or benchmarking of automated deadlift-technique classification methods.MyDeadlift数据集共包含191段标注视频，均为7名具备不同人体测量学特征与力量训练经历的男性参与者完成的硬拉重复动作。该数据集涵盖三类动作类别：正确动作（Correct Movement，印尼语标注：Gerakan Benar；缩写GB；83段视频）、圆背动作（Rounded Back，印尼语标注：Punggung Bungkuk；缩写PB；54段视频）以及膝盖过脚尖动作（Knees Beyond Toes，印尼语标注：Lutut Lebih Jari Kaki；缩写LLK；54段视频）。每段视频对应一次完整的硬拉重复动作，均采用固定侧视角，通过Apple iPhone 13的后置摄像头录制，分辨率为1920×1080像素，帧率为30帧每秒。 本数据集采用受试者划分（subject-wise splitting）协议进行拆分。训练子集包含来自5名参与者的131段视频，测试子集则包含来自2名参与者的60段视频，且每个动作类别均对应20段测试视频。 该数据集仓库还提供了一份经处理的数据表，包含21813条帧级记录。每条记录均涵盖数据划分信息、动作标签、源视频编号、帧序号、膝关节角度、髋关节角度以及背部与竖直方向的夹角。上述生物力学角度均通过MediaPipe Pose提取的人体关键点计算得出。本数据集可复用于人体姿态估计、基于视频的运动动作评估、生物力学动作分析、人体活动识别，以及自动化硬拉动作分类方法的开发与性能基准测试。