Iterative BTreeNet: Unsupervised learning for large and dense 3D point cloud registration

Long Xi, Wen Tang, Tao Xue, Tao Ruan Wan

Abstract: Highlights•A BTreeNet with a novel forward propagation based on the binary tree is proposed.•A novel IBTreeNet is proposed to continuously improve the registration accuracy.•We adopt Chamfer and Earth Mover's Distances as the loss for unsupervised learning.•Our networks are tolerant to partiality and noise without training in such scenes.