Large-scale Point Cloud Semantic Segmentation with Superpoint Graphs
Code (python, pytorch) very soon!
Implementation of the cut pursuit and L0-cut pursuit algorithms to minimize functions with a graph-structured regularizer. The cut pursuit algorithm allows to minimize convex, non-smooth functions regularized by the total variation . The L0 cut pursuit allows to compute piecewise constant approximation of a function defined on a graph when regularized by the weight of the cuts between adjacent constant components.
We provide a c++ implementation as well as a Matlab interface.
Regularization and segmentation framework for point clouds classification (MATLAB)
Provide a benchmark of all methods presented in the paper `A structured regularization framework for spatially smoothing semantic labelings of 3D point clouds`. Only require a ply file and a probabilistic classification to smooth.
An interface for fast partition of point clouds into geometrically simple shapes. I does not provide a one-to-tone instance segmentation of objects, but a sursegmentation in which the clusters are generally semantically homogeneous. As used in the SuperPoint Graph paper.