Portal-Masked Environment Map Sampling · Environment Map Sampling marginal distribution...
Transcript of Portal-Masked Environment Map Sampling · Environment Map Sampling marginal distribution...
Portal-MaskedEnvironment Map Sampling
Benedikt Bitterli 1,2 Jan Novák 2 Wojciech Jarosz 2
1 Walt Disney Animation Studios 2 Disney Research Zurich
Motivation
2
Motivation
3
Environment Map Sampling
4
Environment Map Sampling
shading
point
5
Environment Map Sampling
8
Environment Map Sampling
[Ostromoukhov et al.]
[Kollig and Keller]
9
Environment Map Sampling
occluded
11
Problem Statement
12
Visibility Caching
15
Visibility Caching
[Vorba et al.]
[Georgiev et al.]
[Jensen ‘95]
16
Portal Sampling
portal
18
Portal Sampling
20
Portal Sampling
[Arvo ‘95]
[Ureña et al.]
21
Portal Sampling
23
Ours
24
Environment Map Sampling
25
Environment Map Sampling
marginal distribution
conditional distributions
26
Visible Region
27
Visible Region
28
Visible Region
29
Portal-Masked Environment Map Sampling
Challenges:
1) Visible regions have complex shapes
Marginal/conditional decomposition difficult
30
Portal-Masked Environment Map Sampling
Challenges:
1) Visible regions have complex shapes
Marginal/conditional decomposition difficult
2) Visible regions different for each shading point
Marginal/conditional cannot be precomputed
31
Portal-Masked Environment Map Sampling
Challenges:
1) Visible regions have complex shapes
Marginal/conditional decomposition difficult
2) Visible regions different for each shading point
Marginal/conditional cannot be precomputed
32
Environment Map Parameterizations
Latitude/Longitude Light probe Cube map
33
Environment Map Parameterizations
Latitude/Longitude Light probe Cube map
Visible regions not rectangular in these mappings
34
Environment Map Parameterizations
Latitude/Longitude Light probe Cube map
Visible regions not rectangular in these mappings
Let’s create one in which they are!
Rectified Coordinates
35
Rectified Coordinates
36
Rectified Coordinates
37
Rectified Coordinates
38
Rectified Coordinates
39
Rectified Coordinates
40
Rectified Coordinates
41
Rectified Coordinates
Parametrization is aligned with the portal
42
Rectified Coordinates
Parametrization is aligned with the portal
43
Rectified Coordinates
Need separate environment map per portal44
Rectified Coordinates
46
Rectified Coordinates
47
Rectified Coordinates
48
Rectified Coordinates
49
Rectified Coordinates
50
Rectified Coordinates
51
Rectified Coordinates
52
Rectified Coordinates
53
Rectified Coordinates
54
Rectified Coordinates
55
Rectified Coordinates
56
Rectified Coordinates
57
Rectified Coordinates
58
Portal-Masked Environment Map Sampling
Challenges:
1) Visible regions have complex shapes
Marginal/conditional decomposition difficult
2) Visible regions different for each shading point
Marginal/conditional cannot be precomputed
59
Portal-Masked Environment Map Sampling
Challenges:
1) Visible regions have complex shapes
Marginal/conditional decomposition difficult
2) Visible regions different for each shading point
Marginal/conditional cannot be precomputed
60
Sampling the Visible Region
1) Construct marginal/conditional distribution
2) Invert marginal/conditional distribution
61
Marginal Distribution
62
Marginal Distribution
63
Marginal Distribution
64
Marginal Distribution
Precompute summed-area table instead [Crow ‘84]
65
Sampling the Visible Region
1) Construct marginal/conditional distributions
2) Invert marginal/conditional distributions
66
Precompute summed-area table instead [Crow ‘84]
Inverting the Marginal Distribution
67
Inverting the Marginal Distribution
68
Inverting the Marginal Distribution
69
Inverting the Marginal Distribution
70
Inverting the Marginal Distribution
71
Inverting the Marginal Distribution
72
Sampling the Visible Region
1) Construct marginal/conditional distributions
2) Invert marginal/conditional distributions
73
Precompute summed-area table instead [Crow ‘84]
Use bisection algorithm for logarithmic complexity
Results
74
Results
Comparison of:
a) Environment map sampling
b) Portal solid angle sampling [Ureña et al. 2013]
c) Multiple importance sampling of a) and b) [Veach 1998]
d) Ours
at equal sample count (8spp)
All methods combined with 1 BRDF sample using MIS
75
77
78
79
80
Reference
RMSE 81
Reference
RMSE
Env. Map
0.23
8 spp
82
Reference
RMSE
Env. Map
0.23
Portal S.Angle
3.31
8 spp
83
Reference
RMSE
Env. Map
0.23
Portal S.Angle
3.31
MIS
0.26
8 spp
84
Reference
RMSE
Env. Map
0.23
Portal S.Angle
3.31
MIS
0.26
8 spp
Ours
0.16 85
87
88
89
Reference
RMSE 90
Reference
RMSE
Env. Map
0.59
Portal S.Angle
0.74
8 spp
92
Reference
RMSE
Env. Map
0.59
Portal S.Angle
0.74
MIS
0.60
8 spp
93
Reference
RMSE
Env. Map
0.59
Portal S.Angle
0.74
MIS
0.60
Ours
0.34
8 spp
94
More Results
Please see supplementary material
95
By the numbers
Equal time:
• 33% less shadow rays than Env. Map sampling
Equal quality:
• 2x – 8x speedup w.r.t. Env. Map sampling
• 2x – 5x speedup w.r.t. Env Map/Portal MIS
96
Area lights with directional profile
97
Area lights with directional profile
98
Uniform Solid Angle Sampling Ours
10x speedup
Conclusion
Take away from this talk:
• Can importance sample visibility x Env. Map
• Need single (precomputed) SAT per portal
• Easy to implement!
100
Conclusion
Take away from this talk:
• Can importance sample visibility x Env. Map
• Need single (precomputed) SAT per portal
• Easy to implement!
101
Conclusion
Take away from this talk:
• Can importance sample visibility x Env. Map
• Need single (precomputed) SAT per portal
• Easy to implement!
102
Conclusion
Take away from this talk:
• Can importance sample visibility x Env. Map
• Need single (precomputed) SAT per portal
• Easy to implement!
103
Limitations
104
• Needs a portal to work
Limitations
105
• Needs a portal to work
• Only considers visibility through a rectangle
Limitations
106
• Needs a portal to work
• Only considers visibility through a rectangle
Future Work
Incorporate BRDF:
Combine with product sampling techniques
107
[Burke et al.] [Clarberg et al.] [Cline et al.]
Thanks!
Acknowledgements: • Environment maps: Bernhard Vogl, http://openfootage.net
• Scenes: Jay Hardy, Wayne
108
Bonus Slides
109
Sampling multiple portals
110
Sampling multiple portals
111
Sampling multiple portals
112
Sampling multiple portals
113
Scene Sample Allocation