Complete point clouds using unsupervised learning by the Sinkhorn algorithm with UDPReg
Complete point clouds using unsupervised learning by the Sinkhorn algorithm with UDPReg
Unsupervised Deep Probabilistic Approach for Partial Point Cloud Registration
arXiv paper abstract https://arxiv.org/abs/2303.13290
arXiv PDF paper https://arxiv.org/pdf/2303.13290.pdf
GitHub https://github.com/gfmei/UDPReg
Deep point cloud registration methods face challenges to partial overlaps and rely on labeled data … propose UDPReg, an unsupervised deep probabilistic registration framework for point clouds with partial overlaps.
… first adopt a network to learn posterior probability distributions of Gaussian mixture models (GMMs) from point clouds.
To handle partial point cloud registration … apply the Sinkhorn algorithm to predict the distribution-level correspondences under the constraint of the mixing weights of GMMs.
To enable unsupervised learning … design three distribution consistency-based losses: self-consistency, cross-consistency, and local contrastive.
… self-consistency loss is formulated by encouraging GMMs in Euclidean and feature spaces to share identical posterior distributions … local contrastive loss facilitates … extract discriminative local features.
… UDPReg achieves competitive performance on the 3DMatch/3DLoMatch and ModelNet/ModelLoNet benchmarks.
Stay up to date. Subscribe to my posts https://morrislee1234.wixsite.com/website/contact
Web site with my other posts by category https://morrislee1234.wixsite.com/website
LinkedIn https://www.linkedin.com/in/morris-lee-47877b7b
