ReLift uses pretrained 2D trackers for dense 3D tracking via lifted trajectory repair

Dense 3D tracking from monocular video needs point correspondences, depth estimation, and trajectory inference during occlusion; joint learning is demanding due to scarce 4D supervision, but strong pretrained 2D trackers and 3D geometry models provide a foundation where lifted trajectories with interpolated occluded segments rival dedicated 3D trackers using the same geometry.