Ioannis Gkioulekas 1 Bei Xiao 2 Shuang Zhao 3 Edward Adelson 2 Todd Zickler 1 Kavita Bala 3 1 Harvard 3 Cornell 2 MI Τ 1 Translucency is everywhere food skin jewelry ID: 425456
Download Presentation The PPT/PDF document "Understanding the role of phase function..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Understanding the role of phase function in translucent appearance
Ioannis
Gkioulekas
1
Bei Xiao2
Shuang Zhao3
Edward Adelson2
Todd Zickler1
Kavita Bala3
1Harvard
3Cornell
2
MIΤ
1Slide2
Translucency is everywhere
food
skin
jewelry
architecture
2Slide3
Subsurface scattering
radiative
transfer equation
Chandrasekhar 1960
phase function p
absorption coefficient
σ
a
extinction coefficient
σ
t
3
isotropic
incident direction
outgoing direction
(
λ)
(
λ)
(
λ)Slide4
Phase function is important
thick parts (diffusion)
thin parts
4Slide5
Common phase functions
single-parameter family:
Henyey
-Greenstein (HG) lobes
5
Henyey
and Greenstein 1941
average cosine
Slide6
What can we represent with HG?
microcrystalline wax
6
marble
white jade
Jensen 2001Slide7
Henyey-Greenstein is not enough
s
oap
microcrystalline wax
photo
HG
s
etup
7Slide8
Goals
8
expanded phase function space
role in translucent appearance
?
?Slide9
Expanded phase function space
single-parameter family:
Henyey
-Greenstein (HG) lobes
9
average cosine
second moment
von
Mises
-Fisher (
vMF
) lobes
single-parameter family:
Slide10
Expanded phase function space
s
oap
microcrystalline wax
photo
HG
s
etup
vMF
10Slide11
Expanded phase function space
single-parameter family:
Henyey
-Greenstein (HG) lobes
von
Mises
-Fisher (
vMF
) lobes
single-parameter family:
Linear mixtures:
HG + HG
HG +
vMF
vMF
+
vMF
11
Slide12
f
( )
f( )
Redundant phase
function space
≈
≠
12
≈Slide13
Related work
13
Fleming and
Bülthoff 2005, Motoyoshi
2010
Pellacini et al. 2000, Wills et al. 2009
many perceptual cuesdo not study phase function
gloss perception
much smaller spaceNgan
et al. 2006
gloss perceptionnavigation of appearance spaceSlide14
Our approach
1. Computational
processing
2
. Psychophysical validation
3. Analysis of results
i
mage-driven analysis
tractable experiment
visualization, perceptual parameterization
14Slide15
Scene design
mostly low-order scattering
mostly high-order scattering
side-lighting
thick body and base
thin parts and fine details
15Slide16
von
Mises
-Fisher (
vMF
) lobes
Linear mixtures:
HG + HG
HG +
vMF
Henyey
-Greenstein (HG) lobes
Expanded phase function space
16
sample 750+ phase functions
3000 machine hours
750+ HDR imagesSlide17
Psychophysics
Paired-comparison experiments
Hmm, left
17Slide18
Psychophysics
750 images = 200 million comparisons
18Slide19
d( , )
ǁ - ǁ
Image-driven analysis
≈
19Slide20
two-dimensional appearance space
two-dimensional embedding
Computational processing
750 HDR images
ǁ - ǁ
multidimensional scaling
20
≈Slide21
Our approach
1. Computational
processing
2. Psychophysical validation
i
mage-driven analysis
tractable experiment
21
3. Analysis of results
visualization, perceptual parameterizationSlide22
40 representative images
Psychophysical validation
ǁ - ǁ
clustering
two-dimensional appearance space
22Slide23
Psychophysical validation
750 phase functions = 200 million comparisons
40
phase functions = 30,000 comparisons
23Slide24
computational embedding
Psychophysical validation
24
≈
perceptual embedding
use computational embedding as proxy for psychophysics
generalize to all 750 images
(non-metric MDS on psych. data)
(MDS using image metrics)Slide25
computational embedding
Psychophysical validation
25
≈
perceptual embedding
use computational embedding as proxy for psychophysics
generalize to all 750 images
(non-metric MDS on psych. data)
(MDS using image metrics)Slide26
Our approach
1. Computational
processing
2. Psychophysical validation
i
mage-driven analysis
tractable experiment
26
3. Analysis of results
visualization, perceptual parameterizationSlide27
What we know so far
translucent appearance space
two-dimensional
perceptual
consistent across
variations of material, shape,
illumination
27see paper for: 5000+ images, 9 more computational
embeddings, 2 more psychophysical experiments including backlighting, analysis and statisticsSlide28
Moving around the space
28Slide29
Moving around the space
moving vertically
m
ore diffused appearance
29Slide30
Moving around the space
moving vertically
m
ore diffused appearance
30Slide31
Moving around the space
moving horizontally
m
ore glass-like appearance
31Slide32
Moving around the space
moving horizontally
m
ore glass-like appearance
32Slide33
we can move anywhere
Moving around the space
33Slide34
What can we render with…
single forward lobes
forward + isotropic mixtures
forward + backward mixtures
34Slide35
What can we render with…
single forward lobes
forward + isotropic mixtures
forward + backward mixtures
35Slide36
What can we render with…
marble
white jade
marble
white jade
with
vMF
+
vMF
best approximation
with HG + isotropic
36
≠Slide37
Editing the phase function
move horizontally
move vertically
37
more glass-like
more diffusedSlide38
g
Perceptual parameterization
move vertically
0.8
0.4
0
38
HG:
Slide39
HG:
0.32
Perceptual parameterization
move vertically
0.64
g
2
39
0Slide40
HG:
Perceptual parameterization
40
move vertically
0
g
0.8
0.4
0.32
0.64
g
2
HG:
Slide41
Discussion
41
handling other parameters of appearance:
σt,
σa, color
more general or data-driven phase function models
use in translucency editing and design user interfaces
need to (further) scale up methodology
see our SIGGRAPH Asia 2013 paper!Slide42
Three take-home messages
HG is not enough
expanded space
computation + psychophysics
large-scale perceptual studies
2D appearance spaceuniform parameterization
42
white jade
marbleSlide43
Acknowledgements
Wenzel
Jakob
Bonhams
Funding:NSF NIH
Amazon
white jade
marble
43
http://tinyurl.com/s2013-translucency
Dataset of 5000+ images:Slide44
Computational embeddings
material variation
shape variation
lighting variation
5000+ more HDR imagesSlide45
Scene design
45Slide46
computational embedding
Psychophysical validation
46
≈
perceptual embedding
(non-metric MDS on psych. data)
(MDS using image metrics)Slide47
Computational metrics
L
1
-norm
L
2
-norm
c
ubic rootSlide48
Perceptual image metrics
material variation
shape variation
lighting variationSlide49
Embedding stability
original
perturbation 1
perturbation 2
perturbation 3
perturbation 4
perturbation 5Slide50
Distance metric
MDS
Davis et al. 2007
sample 750+ phase functions
MDSSlide51
Non-metric MDS
Wills et al. 2009
Learning from relative comparisons
non-metric
MDS
d >d
Hmm, left