Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael...

24
Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik Saarbrücken, Germany

Transcript of Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael...

Page 1: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

Efficient Acquisition andRealistic Rendering of Car Paint

Johannes Günther, Tongbo Chen, Michael Goesele,

Ingo Wald, and Hans-Peter Seidel

MPI Informatik

Saarbrücken, Germany

Page 2: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 2

Motivation

Virtual prototyping, car design by computer

Mainly two materials– Glass: ok, physical

properties well known– Car paint: not so easy

Goal:

Realistic appearance of virtual cars, close to reality Phong BRDF: “plastic” look

Page 3: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 3

Introduction & Previous Work Efficient Acquisition

– Measurement Setup– BRDF Representation and Modelling

Realistic Rendering– BRDF Evaluation– Illumination– Simulation of Sparkling

Results Conclusion & Future Work

Outline

Page 4: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 4

Previous Work

BRDF Acquisition [Marschner ‘98, Matusik ‘03]– Image based, automatic fast

Car paint [Ershov ‘01, ‘04]– Complex models, many effects– Not designed for animation context

Illumination by Environment Maps [Debevec ‘98] Realtime Ray Tracing [Wald ’01, ‘04]

Page 5: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 5

Introduction & Previous Work Efficient Acquisition Realistic Rendering Results Conclusion & Future Work

Outline

Page 6: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 6

Measurement Setup

CCD camera white LED

turn tablepainted sphere

Page 7: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 7

Measurement Process

Turn table: rotate light source 180° every 1° At each position: take HDR image

– One view direction, one light direction– Sphere: each pixel different normal

many BRDF sample at once

Time: ca. 30 minutes per target

Page 8: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 8

Targets

Page 9: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 9

Modeling

Use Cook-Torrance BRDF– physically derived (micro facets)– showed to perform well [Ngan EGSR ‘05]

Non-linear fitting Multiple lobes to account for nature of car paints

Page 10: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 10

refle

cta

nce

φ

Modeling

Use Cook-Torrance BRDF– physically derived (micro facets)– showed to perform well [Ngan EGSR ‘05]

Non-linear fitting Multiple lobes to account for nature of car paints

base color

highlight(clear coat)

glitter(flakes)

Page 11: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 11

Introduction & Previous Work Efficient Acquisition Realistic Rendering Results Conclusion & Future Work

Outline

Page 12: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 12

Complex Illumination

HDR Environment Maps for direct illumination Options for BRDF evaluation:

a) Sample Environment Map– Discretize into directional lights [Kollig ‘03, Agarwal ‘03, …]

– Works well for diffuse BRDFs

b) Sample BRDF– Good for specular BRDFs

Decompose car paint BRDF

into diffuse part and

highly specular partre

flect

an

ce

φ

split

hig

hly

sp

ecu

lar

mostly diffuse

car paint BRDF

Page 13: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 13

Sparkles

Prominent feature of metallic paints

Tiny bright spots when viewed from close distance

Caused by mirror-like flakes Reflect light directly to eye

base color

flakes

clear coat

Page 14: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 14

Modeling Flakes

Coherent sparkles during animation Model flakes explicitly (the normal)

(Integrated) sparkles appear as glitter in BRDF Derive statistical flake distribution from fitted glitter lobe

Use procedural normal map Flakes are very small

anti-aliasing by over sampling

Page 15: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 15

Introduction & Previous Work Efficient Acquisition Realistic Rendering Results Conclusion & Future Work

Outline

Page 16: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 16

Model Comparison

Phong ClearCoat™

BRDF table fitted BRDF

Page 17: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 17

Video

Page 18: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 18

Conclusion

Easy-to-build and fast acquisition systemMeasured car paint and measured lighting

environment for convincing car renderingsFrame-to-frame coherent sparkling simulation

Future Work– Extend car paint database– Multi-level methods for sparkles (avoid aliasing)

Page 19: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 19

Project homepage:http://www.mpi-inf.mpg.de/~guenther/carpaint/

Data sets available

Page 20: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 20

Project homepage:http://www.mpi-inf.mpg.de/~guenther/carpaint/

Thank You

Questions?

Page 21: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 21

Page 22: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 22

Performance vs. Quality

640 × 48016 lights

no over sampling12.1 fps

640 × 480128 lights

16 spp1.3 fps

1280 × 9601024 lights

64 spp97 sec

Cluster of 20 dual Opteron 2.5 GHz PCs Vary parameter to tune rendering speed or quality

Page 23: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 23

Offline Rendering

Page 24: Efficient Acquisition and Realistic Rendering of Car Paint Johannes Günther, Tongbo Chen, Michael Goesele, Ingo Wald, and Hans-Peter Seidel MPI Informatik.

November 18, 2005 VMV, Erlangen, Germany 24

The Different Car Paints