Error analysis of estimators that use combinations of stochastic sampling strategies for direct illumination Kartic Subr, Derek Nowrouzezahrai, Wojciech Jarosz, Jan Kautz and Kenny Mitchell Disney Research, University of Montreal, University College London
50
Embed
Kartic Subr , Derek Nowrouzezahrai , Wojciech Jarosz , Jan Kautz and Kenny Mitchell
Kartic Subr , Derek Nowrouzezahrai , Wojciech Jarosz , Jan Kautz and Kenny Mitchell Disney Research, University of Montreal, University College London. Error analysis of estimators that use combinations of stochastic sampling strategies for direct illumination. - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Error analysis of estimators that use combinations ofstochastic sampling strategies for direct illuminationKartic Subr, Derek Nowrouzezahrai, Wojciech Jarosz, Jan Kautz and Kenny Mitchell
Disney Research, University of Montreal, University College London
error of unbiased stochastic estimator = sqrt(variance)
error of deterministic estimator = bias
Variance depends on samps. per estimate ‘N’
estimated value (bins)
Num
ber o
f esti
mat
es
estimated value (bins)N
umbe
r of e
stim
ates
Histogram of 1000 estimates with N =10
Histogram of 1000 estimates with N =50
increasing ‘N’, error approaches bias
N
erro
r
convergence rate of estimator
N
erro
r
log-Nlo
g-er
ror
convergence rate = slope
comparing estimators
log-
erro
r
log-N
Estimator 2
Estimator 1
Estimator 3
the “better” estimator depends on application
log-
erro
r
log-N
Estimator 2
Estimator 1
real
-tim
e
Offl
ine
sample budget
typical estimators (anti-aliasing)
log-
erro
r
log-N
random MC (-0.5)
MC with jittered sampling (-1.5)QMC (-1)
randomised QMC (-1.5)
MC with importance sampling (-0.5)
What happens when strategies are combined?
Z X Y= +
combined estimate
single estimate using estimator 1
single estimate using estimator 2
( ) / 2
?
& what about convergence?
What happens when strategies are combined?
Non-trivial, not intuitiveneeds formal analysis
can combination improve convergence or only constant?
we derived errors in closed form…
combinations of popular strategies
Improved convergence by combining …
Strategy A Strategy B Strategy C
New strategy D Observed
convergence of D is better than
that of A, B or C
exciting result!
Strategy A Strategy B Strategy C
New strategy D
jittered antithetic importance
Observedconvergence of D
is better than that of A, B or C
related work
Correlated and antithetic sampling
Combining variance reduction
schemes
Monte Carlo sampling
Variance reduction
Quasi-MC methods
Intuition(now)
Formalism(suppl. mat)
recall combined estimator
Z X Y= +
combined estimate
single estimate using estimator 1
single estimate using estimator 2
( ) / 2
applying variance operator
Z X Y= +
combined estimate
single estimate using estimator 1
single estimate using estimator 2
V ( ) V ( ) V ( )( ) / 4
+ 2 cov ( , ) /4X Y
variance reduction via negative correlation
Z X Y= +V ( ) V ( ) V ( )( ) / 4
+ 2 cov ( , ) /4X Y
combined estimate
single estimate using estimator 1
single estimate using estimator 2
best case is when X and Y have a correlation of -1
“antithetic” estimates yield zero variance!
Z X Y= +V ( ) V ( ) V ( )( ) / 4
+ 2 cov ( , ) /4X Y
combined estimate
single estimate using estimator 1
single estimate using estimator 2
antithetic estimates vs antithetic samples?
X YX = f(s) Y = f(t)
x
f(x)
s t
X
Y
antithetic estimates vs antithetic samples?
x
f(x)
s t
X
Y
if t=1-s, corr(s,t)=-1- (s,t) are antithetic samples- (X,Y) are not antithetic estimates unless f(x) is linear!- worse, cov(X,Y) could be positive and increase overall variance
antithetic sampling within strata
x
f(x)
s t
?
integrand not linear within many strata• But where linear, variance is close to zero
• As number of strata is increased, more benefit– i.e. if jittered, benefit increases with ‘N’– thus affects convergence
• Possibility of increased variance in many strata
• Improve by also using importance sampling?
review: importance sampling as warp
x
f(x)
g(x)
uniform
Ideal case: warped integrand is a constant
with antithetic: linear function sufficient
x
with
with
with
with stratification + antithetic: piece-wise linear is sufficient
x
Warped integrand
summary of strategies• antithetic sampling
– zero variance for linear integrands (unlikely case)
• stratification (jittered sampling)– splits function into approximately piece-wise linear
• importance function– zero variance if proportional to integrand (academic case)
summary of combination• Stratify
• Find importance function that warps into linear func.