LightTrack implements a top-down human pose tracking pipeline using deep learning pose estimators. It performs multi-person pose estimation across video frames and applies data association techniques to track individuals over time. The framework supports live webcam input, arbitrary video files, and includes training capabilities for custom models.