Search papers, labs, and topics across Lattice.
This paper introduces Sinkhorn-CPD, an enhancement of the Coherent Point Drift (CPD) algorithm for rigid point cloud registration that addresses the limitations of CPD's target-side marginal constraint, which struggles with outliers and partial overlaps. By integrating dual Kullback-Leibler penalties, Sinkhorn-CPD effectively discards outliers and formulates the problem as a fully unbalanced entropic optimal transport issue, solvable via generalized Sinkhorn iterations. Experimental results demonstrate that Sinkhorn-CPD achieves state-of-the-art accuracy and robustness across various benchmarks, outperforming existing methods in challenging scenarios.
Sinkhorn-CPD achieves state-of-the-art point cloud registration accuracy while automatically handling outliers and partial overlaps without manual tuning.
Coherent Point Drift (CPD) is widely used for rigid point cloud registration because of its soft correspondences and closed-form parameter updates. However, CPD's target-side marginal constraint forces every observation, including outliers, to receive exactly unit probability mass. This assumption degrades registration accuracy under heavy outliers and partial overlap. Optimal transport (OT) methods can handle missing mass through unbalanced formulations, but require hand-tuned annealing schedules. In this paper, we propose Sinkhorn-CPD, which replaces CPD's target-side marginal constraint with dual Kullback-Leibler penalties, allowing the algorithm to discard outliers on both sides. The resulting formulation is a fully unbalanced entropic optimal transport problem, which can be efficiently solved by generalized Sinkhorn iterations. Moreover, Sinkhorn-CPD preserves the closed-form Procrustes and variance updates of CPD. In our method, the variance sigma^2 plays the role of the entropic regularization parameter, which induces an automatic annealing schedule from diffuse to sharp correspondences without manual temperature tuning. Experiments on synthetic, cross-category, and scan-to-CAD benchmarks show that Sinkhorn-CPD achieves state-of-the-art accuracy, with strong robustness to outliers and partial overlap.
