Validation of a handheld smartphone-based markerless computer vision algorithm with estimated groundline for equine gait analysis

This dataset contains stride- and trial-level measurements of vertical displacement symmetry (MaxDiff and MinDiff) for eye, withers, back, and croup keypoints, recorded in 58 horses trotting on a straight line and 24 horses trotting on a circle. Data include outputs from the markerless computer vision algorithm (RealHorse®) and the reference 3D optical motion capture system (Qualisys®), with processed error metrics (MSE and MAE). The dataset has been curated to exclude back-up camera B recordings and observations without valid reference data (bad_QS), and further stratified by movement type (straight line or circle) and keypoint detected (eye, withers, back, and croup). All original data is included.