Out-of-Core Construction of Sparse Voxel Octrees

Supplemental Material

Jeroen Baert, Ares Lagae and Philip Dutre

Department of Computer Science, KU Leuven, Belgium∗

Abstract

This document report contains supplemental material for our paperOut-of-Core Construction of Sparse Voxel Octrees.

1 Additional Results

In Tab. 1 we show additional results of our method. These resultsconfirm that our out-of-core algorithm can construct sparse voxeloctrees from triangle meshes using only a fraction of the memoryrequired by an in-core algorithm in roughly the same time, and thatour out-of-core algorithm can also handle extremely large trianglemeshes.

∗e-mail: {jeroen.baert, ares.lagae, philip.dutre}@cs.kuleuven.be

XYZ RGB Dragon Nature St.Matthew

(a) method parameters

grid resolution 20483

20483

40963

grid size 8 GB 8 GB 64 GB

available memory 128 MB 1 GB 8 GB 128 MB 1 GB 8 GB 1 GB

(b) input triangle mesh statistics

# triangles 7.2 M 41.3 M 372.76 M

size 248 MB 1.45 GB 13.1 GB

(c) voxelization process statistics

# partitions / subgrids 64 8 1 64 8 1 64

(c.1) partitioning subprocess statistics

time, IO, input 0.22 s 0.21 s 0 s 1.21 s 1.20 s 0 s 248.66 s

time, partitioning 1.63 s 0.30 s 0 s 11.24 s 1.80 s 0 s 74.43 s

time, IO, output 0.08 s 0.07 s 0 s 0.52 s 0.50 s 0 s 39.08 s

time, misc 0.28 s 0.24 s 0 s 1.43 s 1.72 s 0 s 12.20 s

time, total 2.21 s 0.84 s 0 s 14.41 s 5.22 s 0 s 410.38 s

(c.2) intermediate triangle mesh partitions statistics

# empty part. / subg. 38 0 0 31 0 0 48

(c.3) voxelization subprocess statistics

time, IO, input 0.22 s 0.23 s 0.22 s 1.21 s 1.23 s 4.38 s 10.47 s

time, voxelization 9.38 s 14.99 s 18.08 s 204.68 s 209.12 s 211.94 s 97.50 s

time, IO, output 0.27 s 0.23 s 0.28 s 2.85 s 3.05 s 2.90 s 1.27 s

time, misc 0.41 s 0.61 s 0.37 s 2.82 s 3.13 s 2.80 s 13.43 s

time, total 10.29 s 16.07 s 18.97 s 211.58 s 216.55 s 222.03 s 122.68 s

time, total 12.51 s 16.92 s 18.97 s 225.99 s 221.78 s 222.03 s 533.06 s

(d) intermediate high-resolution 3D voxel grid statistics

# non-empty cells 3.9 M 44.17 M 18.60 M

sparseness 99.96% 99.50% 99.80 %

size 31 MB 773 MB 145 MB

(e) sparse voxel octree construction process statistics

time, IO, input 0.26 s 3.04 s 4.41 s

time, SVO const. 0.87 s 7.80 s 3.98 s

(unoptimized version) (471.79 s) (N/A) (N/A)

time, IO, output 0.93 s 9.64 s 4.32 s

time, misc 0.31 s 3.37 s 1.42 s

time, total 2.38 s 23.87 s 14.13 s

(f) output sparse voxel octree statistics

# nodes 5.71 M 59.14 M 26.22 M

size 131 MB 1.38 GB 614 MB

(g) total time 14.90 s 19.31 s 21.36 s 249.86 s 245.65 s 245.90 s 547.20 s

(h) visualization and rendering of sparse voxel octree

Table 1: Additional results. Our out-of-core algorithm can construct sparse voxel octrees from triangle meshes using only a fraction of thememory required by an in-core algorithm in roughly the same time, and can also handle extremely large models (see Sec. 1).