In the following, we show the result of exwayz_subsampling.exe on the following dense point cloud
0 )The input point cloud is inserted in a grid with a given resolution r . A voxel can only be occupied by the first point falling in it (which is quite arbitrary), resulting in point cloud with 1 point per voxel of side r . It doesn’t guarantee a minimum distance between points for instance, or any regularity in the spatial subsampling, but this method is very fast.
Result of exwayz_subsampling.exe with commandline options --mode 0 -r 0.2
1 )The first point of the input point cloud is tagged as “kept”, then all its neighbors in a radius r are tagged a “dismissed”. We iterate on the whole point cloud this way, skipping the described tagging operation if the current point is already tagged as “dismissed”. This process guarantees a minimum distance of r between points in the resulting point cloud.
Result of exwayz_subsampling.exe with commandline options --mode 1 -r 0.2
2 )The indices of the point cloud are randomly shuffled and the n first points are kept as the remaining points.
