position-free monte carlo simulation for arbitrary layered...

1
Position-Free Monte Carlo Simulation for Arbitrary Layered BSDFs Yu Guo 1 , Miloš Hašan 2 and Shuang Zhao 1 1 University of California, Irvine 2 Autodesk Inc. ACM Transactions on Graphics (SIGGRAPH Asia 2018), 37(6), 2018 Introduction In this paper, we introduced a new BSDF model to capture the appearance of layered materials. Inside the evaluation and sampling routines of the layered BSDF, we run a Monte Carlo simulation of light transport within flat slabs. This is substantially faster than explicitly constructing the layer geometry, but also allows constructing light transport paths that would not easily be available to a generic light transport algorithm, due to our new position-free path formulation. Within this framework, we introduced unbiased Monte Carlo techniques analogous to a forward path tracer with next event estimation (NEE) and a fully bidirectional estimator. We demonstrated the capabilities of our solution on a number of examples, featuring multiple layers with surface and volumetric scattering, surface and phase function anisotropy, and spatial variation in all parameters. This leads to the first BSDF layering solution that offers unbiased accuracy and full flexibility in setting the layer properties. Assumption Small displacement assumption: Since the geometric thickness h of the layer is small, we assume the displacements (e.g., Δx) of light’s entrance and departure locations can be neglected. Methods BSDF sampling: Forward path tracing in flat slab configuration BSDF evaluation: Unidirectional estimator (c) (d) (e) Bidirectional estimator Pdf estimation: 45° 90° 135° 180° 225° 270° 315° Groundtruth pdf 0.0 0.1 0.2 0.3 0.4 0.5 0.6 45° 90° 135° 180° 225° 270° 315° Unbiased pdf 0.0 0.1 0.2 0.3 0.4 0.5 0.6 45° 90° 135° 180° 225° 270° 315° Approximate pdf 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Comparison Equal-time comparison: 98 spp 35 spp 280 spp 56 spp 26 spp 60 spp 25 spp 80 spp 15 spp 19 spp Reference PT BDPT MLT Our uni-dir Our bi-dir Comparison to previous work: not available expensive precomputation not available not available Ours [Zeltner 2018] [Belcour 2018] Comparison to volumetric cloth: Volume rendering Our BSDF + fiber-direction map More results Transmission (Spectrally varying optical densities): Transmission (Different phase function): HG (g =0.9) HG (g =0.99) vMF (κ = 10) vMF (κ = 100) Anisotropic media within layers: Satin Twill Velvet Multi-layer BSDF: References [1] Tizian Zeltner and Wenzel Jakob. The layer laboratory: A calculus for additive and subtractive composition of anisotropic surface reflectance. ACM Trans. Graph., 2018. [2] Laurent Belcour. Efficient rendering of layered materials using an atomic decomposition with statistical operators. ACM Trans. Graph., 2018.

Upload: others

Post on 19-Sep-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Position-Free Monte Carlo Simulation for Arbitrary Layered ...yug10/webpage/pdf/2018SigAsia...Position-Free Monte Carlo Simulation for Arbitrary Layered BSDFs Yu Guo1, Miloš Hašan2

Position-Free Monte Carlo Simulation for Arbitrary Layered BSDFsYu Guo1, Miloš Hašan2 and Shuang Zhao1

1University of California, Irvine 2Autodesk Inc.

ACM Transactions on Graphics (SIGGRAPH Asia 2018), 37(6), 2018

Introduction

In this paper, we introduced a new BSDFmodel to capture the appearanceof layered materials. Inside the evaluation and sampling routines of thelayered BSDF, we run a Monte Carlo simulation of light transport withinflat slabs. This is substantially faster than explicitly constructing the layergeometry, but also allows constructing light transport paths that wouldnot easily be available to a generic light transport algorithm, due to ournew position-free path formulation.

Within this framework, we introduced unbiased Monte Carlo techniquesanalogous to a forward path tracer with next event estimation (NEE) anda fully bidirectional estimator. We demonstrated the capabilities of oursolution on a number of examples, featuring multiple layers with surfaceand volumetric scattering, surface and phase function anisotropy, andspatial variation in all parameters. This leads to the first BSDF layeringsolution that offers unbiased accuracy and full flexibility in setting thelayer properties.

Assumption

Small displacement assumption: Since the geometric thickness hof the layer is small, we assume the displacements (e.g., ∆x) of light’sentrance and departure locations can be neglected.

MethodsBSDF sampling: Forward path tracing in flat slab configurationBSDF evaluation:

(c) (d) (e)(a) (b)Unidirectional estimator

(c) (d) (e)(a) (b)Bidirectional estimator

Pdf estimation:

45°

90°

135°

180°

225°

270°

315°

Groundtruth pdf

0.0 0.1 0.2 0.3 0.4 0.5 0.6

45°

90°

135°

180°

225°

270°

315°

Unbiased pdf

0.0 0.1 0.2 0.3 0.4 0.5 0.6

45°

90°

135°

180°

225°

270°

315°

Approximate pdf

0.0 0.1 0.2 0.3 0.4 0.5 0.6

ComparisonEqual-time comparison:

98 spp 35 spp 280 spp 56 spp 26 spp

60 spp 25 spp 80 spp 15 spp 19 sppReference PT BDPT MLT Our uni-dir Our bi-dir

Comparison to previous work:

not available

expensiveprecomputation

not available not available

Ours [Zeltner 2018] [Belcour 2018]

Comparison to volumetric cloth:

Volume rendering

Our BSDF + fiber-direction map

More resultsTransmission (Spectrally varying optical densities):

Transmission (Different phase function):

HG (g = 0.9) HG (g = 0.99) vMF (κ = 10) vMF (κ = 100)

Anisotropic media within layers:Satin Twill Velvet

Multi-layer BSDF:

References[1] Tizian Zeltner and Wenzel Jakob.

The layer laboratory: A calculus for additive and subtractive composition ofanisotropic surface reflectance.ACM Trans. Graph., 2018.

[2] Laurent Belcour.Efficient rendering of layered materials using an atomic decomposition withstatistical operators.ACM Trans. Graph., 2018.