Precomputed Local Radiance Transfer for Real-time Lighting Design Anders Wang Kristensen Tomas...

Post on 19-Dec-2015

216 views 1 download

Tags:

Transcript of Precomputed Local Radiance Transfer for Real-time Lighting Design Anders Wang Kristensen Tomas...

1

Precomputed Local Radiance Transfer for Real-time Lighting Design

Anders Wang KristensenTomas Akenine-Moller

Henrik Wann JensenSIGGRAPH ‘05

Presented by Shao-Ti Lee

2010/04/08

2

OutlineIntroductionRelated WorkConstructing the Light CloudCompressing Surface RadianceReal-time RelightingResultsConclusion

3

IntroductionRelated WorkConstructing the Light CloudCompressing Surface RadianceReal-time RelightingResultsConclusion

4

IntroductionMAIN IDEA

◦Unstructured light cloud.

5

IntroductionMAIN IDEA

◦Unstructured light cloud.

6

IntroductionFEATURES

◦Local, not distant illumination. Accurately represent incident radiance on

different parts of the model.

◦Light positions and intensities are changeable.

◦Lights can be added or removed.◦Materials can be glossy.◦Camera is fully dynamic.◦Scene is assumed to be static.

7

IntroductionRelated WorkConstructing the Light CloudCompressing Surface RadianceReal-time RelightingResultsConclusion

8

Related WorkPrecompute the exitant radiance

[Wood et al. 2000; Chen et al. 2002]◦Basis: sphere harmonics.

9

Related WorkSphere harmonics.

ApproximatedExitantRadiance

ApproximatedExitant Radiancein the SH basis

SH Basis

10

Related WorkClustered PCA(Principle

Component Analysis) [Sloan et al. 2003]◦For geometry data representation

compression. ClusteringEx. K-Means

PCA

11

IntroductionRelated WorkConstructing the Light CloudCompressing Surface RadianceReal-time RelightingResultsConclusion

12

Constructing the Light Cloud

Before doing this, first divide the geometry into a set of discrete zones using a simple top-down partitioning algorithm.

13

Constructing the Light Cloud

Use a two-stage approach to construct the light cloud.◦Stage 1 Make a sufficiently dense

uniform sampling of the region of interest.

◦Stage 2 The light cloud is then simplified by clustering similar lights using a bottom-up clustering algorithm.

14

Constructing the Light Cloud

This is for point lights.

15

Constructing the Light Cloud

This is for diffuse surface.

16

Constructing the Light Cloud

A C

D E F

G H I A B C D E F G H I

Cluster List

Priority queue sorted by △jk

△AB △AD △BD △CE …..

S

S

Compute △SC , △SD , △SE , △SF , △SG ,△SH , △SI and re-sort the priority queue to end a loop and stop until △jk ≧1 for all j, k

B

17

IntroductionRelated WorkConstructing the Light CloudCompressing Surface RadianceReal-time RelightingResultsConclusion

18

Compressing Surface Radiance

Before compression◦Per vertex matrix X , with np = nb x

nl elements per color channel. nb : The number of SH bases.

nl : The number of lights.

After compression using CPCA

ClusterMean

PCA vectors(bases)

weights

19

IntroductionRelated WorkConstructing the Light CloudCompressing Surface RadianceReal-time RelightingResultsConclusion

20

Real-time RelightingOrganize the local lights in a kd-

tree and locate the m nearest lights.

21

Real-time RelightingTo avoid undesirable popping

effects, set the weight for each pre-computed local light to

After computing all weights, normalization is used. And all weights are multiplied by the power of the light at l.

22

Real-time RelightingVisibility problem

Solution: Use ray tracing with a few rays.

Cannot assign weights!

23

Real-time RelightingDiscontinuity Problem

Suddenly disappear/appeardue to occlusion!

Solution:Smoothly fade out/in lights over time, but that now have become obscured by geometry.

24

Real-time RelightingComputing Exitant Radiance

For each cluster, xm and bi are constant.

25

Real-time RelightingReconstruct the vector of SH

coefficients representing exitant radiance in a vertex program by evaluating

◦Where is the variable to ensure that we get correct blending at the borders between clusters and between zones.

◦Recall that

26

IntroductionRelated WorkConstructing the Light CloudCompressing Surface RadianceReal-time RelightingResultsConclusion

27

Results

28

Results

29

Results

Left: Method of authors.Right: Ray tracing with per pixel lighting.

30

Results

31

IntroductionRelated WorkConstructing the Light CloudCompressing Surface RadianceReal-time RelightingResultsConclusion

32

ConclusionThe system handles indirect

illumination efficiently for models with more than 100,000 triangles.

Future Work◦Soft shadow◦Spotlight