Datasets
GolfDB: A Video Database for Golf Swing Sequencing
GolfDB is a benchmark database for golf swing sequencing, consisting of 1400 high-quality golf swing videos. Each video is labeled with event frames, bounding boxes, player information (name and sex), club type, and view type. The database was introduced alongside SwingNet, a lightweight deep neural network for detecting key events in golf swings. GolfDB enables consistent evaluation of golf swing sequencing performance and facilitates golf swing analysis through automated event detection. The database supports research in computer vision, biomechanics, and sports analytics for golf.
Read Paper → (opens in a new tab)CaddieSet: A Golf Swing Dataset with Human Joint Features and Ball Information
CaddieSet is a comprehensive golf swing dataset that includes joint information and ball trajectory data from single shots. The dataset extracts joint information from swing videos by segmenting them into eight swing phases using computer vision-based approaches. Based on expert golf domain knowledge, CaddieSet defines 15 key metrics that influence golf swing performance, enabling quantitative analysis of the relationship between swing posture and ball trajectory. The dataset supports interpretable models for predicting ball trajectories and provides swing feedback that is quantitatively consistent with established domain knowledge. CaddieSet offers new insights for golf swing analysis in both academic research and sports industry applications.
Read Paper → (opens in a new tab)SportsPose: A Dynamic 3D Sports Pose Dataset
SportsPose is a large-scale 3D human pose dataset consisting of highly dynamic sports movements. With more than 176,000 3D poses from 24 different subjects performing 5 different sports activities, SportsPose provides a diverse and comprehensive set of 3D poses that reflect the complex and dynamic nature of sports movements. The dataset has been quantitatively evaluated by comparing poses with a commercial marker-based system, achieving a mean error of 34.5 mm across all evaluation sequences. SportsPose contains more movement in extremum joints (wrists and ankles) than Human3.6M and 3DPW datasets, indicating more dynamic movements. The dataset is valuable for developing and evaluating models for sports performance analysis and injury prevention.
Read Paper → (opens in a new tab)MoVi: A Large Multipurpose Motion and Video Dataset
MoVi is a comprehensive human Motion and Video dataset containing 60 female and 30 male actors performing a collection of 20 predefined everyday actions and sports movements, plus one self-chosen movement. The dataset was recorded in five capture rounds using different hardware systems, including optical motion capture, video cameras, and inertial measurement units (IMU). In total, MoVi contains 9 hours of motion capture data, 17 hours of video data from 4 different viewpoints (including one hand-held camera), and 6.6 hours of IMU data. The dataset includes state-of-the-art estimates of skeletal motions and full-body shape deformations, making it valuable for sports biomechanics research, character animation, and movement analysis applications.
Read Paper → (opens in a new tab)