Search papers, labs, and topics across Lattice.
This paper classifies eight baseball pitch types using only monocular 3D pose sequences extracted from broadcast video, achieving 80.4% accuracy on a dataset of 119,561 pitches. The pipeline combines a diffusion-based 3D pose estimation backbone, automatic event detection, biomechanical feature extraction, and gradient-boosted classification. Feature importance analysis reveals that upper-body kinematics, particularly wrist position and trunk lateral tilt, are most predictive, while grip-defined variants are not separable from pose alone, establishing a performance ceiling.
You can predict a baseball pitch type with 80% accuracy just by watching the pitcher's body, but don't expect to distinguish fastball variants without seeing the ball.
How much can a pitcher's body reveal about the upcoming pitch? We study this question at scale by classifying eight pitch types from monocular 3D pose sequences, without access to ball-flight data. Our pipeline chains a diffusion-based 3D pose backbone with automatic pitching-event detection, groundtruth-validated biomechanical feature extraction, and gradient-boosted classification over 229 kinematic features. Evaluated on 119,561 professional pitches, the largest such benchmark to date, we achieve 80.4\% accuracy using body kinematics alone. A systematic importance analysis reveals that upper-body mechanics contribute 64.9\% of the predictive signal versus 35.1\% for the lower body, with wrist position (14.8\%) and trunk lateral tilt emerging as the most informative joint group and biomechanical feature, respectively. We further show that grip-defined variants (four-seam vs.\ two-seam fastball) are not separable from pose, establishing an empirical ceiling near 80\% and delineating where kinematic information ends and ball-flight information begins.
