Self Supervised Scene Flow Estimation

Self-supervised scene flow estimation is a method that predicts the motion of each point using only LiDAR point cloud data, aiming to learn correspondences and motion fields between point clouds through unlabeled data. This technique is widely applied in Argoverse 2.0 and the 2024 Scene Flow Challenge, playing a crucial role in enhancing environmental perception for autonomous driving and robotic navigation.