Voxel51/qualcomm-exercise-video-dataset-benchmark

QEVD-FIT-COACH-Benchmark数据集包含74个锻炼视频会话，带有实时AI健身教练反馈。每个会话时长约2.5-3分钟，展示单个参与者进行4-5种不同锻炼，并接收AI健身教练的持续反馈。数据集包含反馈事件的时间标注，包括锻炼转换、姿势纠正、鼓励和重复计数。所有反馈都与视频帧对齐，并包含精确的时间信息。视频格式为MP4（H.264/AVC编解码器，640x360分辨率，30 FPS），附带JSON标注。数据集适用于视频理解研究、健身AI开发、人机交互和计算机视觉应用。

The QEVD-FIT-COACH-Benchmark dataset contains 74 workout video sessions with real-time AI fitness coaching feedback. Each session is approximately 2.5-3 minutes long and features a single participant performing 4-5 different exercises with continuous feedback from an AI fitness coach. The dataset includes temporal annotations for feedback events, including exercise transitions, form corrections, encouragement, and repetition counting. All feedback is aligned with video frames and includes precise timing information. Videos are in MP4 format (H.264/AVC codec, 640x360 resolution, 30 FPS) with JSON annotations. The dataset is suitable for video understanding research, fitness AI development, human-computer interaction, and computer vision applications.