Aligned Instance Volume Labels to Point Clouds - JamesDarby345/Volumetric_Instance_to_Mesh GitHub Wiki

The file kdtree_instance_to_point_cloud.py, once pointed to the harmonised/aligned labels from the cube_label_reassignment.py file, or from direct download from the data server, will generate a point cloud, of variable density which can be used for meshing directly or as part of the thaumato pipeline.

Sparser point clouds using larger radius values are more robust to errors in the underlying instance labels, but higher resolution and smaller radius values will make point clouds more similar to the underlying instance labels if they are high quality, resulting in meshes that more closely follow the structure.

The approach is highly parralelisable and should be fast enough for usage towards the autosegmentation prize.

Sparse point cloud Higher sample rate point cloud Dense point cloud All cross cube point clouds of manually labelled cubes Point cloud instance groups