Game Development Reference
In-Depth Information
Scene
Grid
GS
G-buffers
PV
Size
4-floats
Creation
4-floats
128 3
Conference
31.73
0.28
3.2
512 3
(282K tris)
64.67
4.93
128 3
Dragon
198.33
0.18
59
(871K tris)
512 3
-
6.98
128 3
Turbine Blade
265.7
0.14
121
(1.76M tris)
512 3
-
5.37
Hairball
128 3
436.2
0.33
-
(2.88M tris)
320 3
-
4.04
Tab l e 6 . 1 . Voxelization timings (in ms) of various scenes using progressive voxelization
(PV) and the geometry slicing (GS) method of [Gaitatzes et al. 11] with 11 output
vertices. We present the total (injection + cleanup) performance values of our 2D
textures implementation using an injection grid proportional to the volume size, which
is our algorithm's worst case as can be seen from the red plot of Figure 6.8.
Scene
Grid
Hausdorff
Size
( X , Y )
% d H
64 3
0.3289
0.2168
0.1091
-
Bunny
128 3
0.1694
256 3
(69.5K tris)
0.1064
64 3
0.3621
0.2565
0.1289
0.0645
Dragon
128 3
0.1878
256 3
(871K tris)
0.1256
64 3
0.3457
0.2763
0.1424
0.0697
Turbine Blade
128 3
0.1821
256 3
(1.76M tris)
0.1232
Tab l e 6 . 2 . Comparison of a full voxelization. We record the normalized (with respect
to the mesh bounding box diagonal) average Hausdorff distance (percent). Mesh X is
the original mesh to be voxelized and Y is the point cloud consisting of the voxel centers
of the voxelization using PV (column 3) and a geometry-based full scene voxelization
(column 4).
resulting voxelization using the PV method (see column 3). We notice that our
voxelized object (voxel centers) is on average 0.1% different from the original
mesh. In addition, we report the Hausdorff distance between the original mesh
and the geometry-based full scene voxelization of [Gaitatzes et al. 11] (see col-
 
 
Search Nedrilad ::




Custom Search