Agreement Between Manual and Automated Keypoints and Symmetry Metrics in a Markerless Vision-Based Algorithm for Equine Gait Analysis

This repository contains data associated with the study “Agreement Between Manual and Automated Keypoints and Symmetry Metrics in a Markerless Vision-Based Algorithm for Equine Gait Analysis.” The study evaluates the performance of a vision-based deep learning algorithm for detecting anatomical keypoints (eye, withers, croup) in horses during trot under field conditions. The algorithm's accuracy is assessed through comparison with manually annotated reference data, analyzing both frame-level vertical keypoint agreement and stride-level gait symmetry metrics (Maxdiff and Mindiff). A sub-study on manual annotation variability further supports the robustness of the reference data. The dataset supports clinical and research applications in equine gait analysis and lameness detection using accessible smartphone-based tools.