Before calculating the TDA features, a voxel downsampling step is performed on the input point cloud. The default voxel size for downsampling is 0.09 m. Thus, for every 9 cm x 9 cm x 9 cm cube in space, all points are replaced by a single representative point. On May 14th, 2026 I noticed while processing scan date 2/4/2022 that this resulted in downsampled point clouds containing between 3-30 points. I suspect this is insufficient to accurately estimate TDA features.
Before calculating the TDA features, a voxel downsampling step is performed on the input point cloud. The default voxel size for downsampling is 0.09 m. Thus, for every 9 cm x 9 cm x 9 cm cube in space, all points are replaced by a single representative point. On May 14th, 2026 I noticed while processing scan date 2/4/2022 that this resulted in downsampled point clouds containing between 3-30 points. I suspect this is insufficient to accurately estimate TDA features.