Open SvenKratz opened 1 year ago
This is a limitation of the ply decoder that is used in Draco. The point cloud compression itself supports arbitrary number of dimensions. A solution would be to parse the data on your end and initiate the draco::PointCloud
data structure manually using the C++ api.
I'm trying to apply draco to a "Gaussian Splatting" point cloud, which has many more dimensions than just point locations, e.g., here are the header fields:
Draco seems to truncate all fields after the first six, assuming a coords + normals representation. Decoded header:
Are there any plans to make the approach more general to support "exotic" point clouds such as those used by Gaussian splatting?