3D Kinematics Estimation from Video with a Biomechanical Model and Synthetic Training Data

The author proposes a novel biomechanics-aware network that directly outputs 3D kinematics from two input views, trained on synthetic data, outperforming previous methods across multiple datasets.