Anton S. Kaplanyan Karlsruhe Institute of Technology, Germany Path Space Regularization Framework.

Slides:



Advertisements
Similar presentations
Bidirectional Photon Mapping Jiří Vorba Charles University in Prague Faculty of Mathematics and Physics 1.
Advertisements

RECENT ADVANCES IN LIGHT TRANSPORT SIMULATION SOME THEORY & A LOT OF PRACTICE Jaroslav Křivánek Charles University in Prague Marcos Fajardo Solid Angle.
Computer Graphics In4/MSc Computer Graphics Lecture Notes #15 Illumination III View Independent Rendering.
Chunhui Yao 1 Bin Wang 1 Bin Chan 2 Junhai Yong 1 Jean-Claude Paul 3,1 1 Tsinghua University, China 2 The University of Hong Kong, China 3 INRIA, France.
Advanced Computer Graphics
ATEC Procedural Animation Introduction to Procedural Methods in 3D Computer Animation Dr. Midori Kitagawa.
MATHIEU GAUTHIER PIERRE POULIN LIGUM, DEPT. I.R.O. UNIVERSITÉ DE MONTRÉAL GRAPHICS INTERFACE 2009 Preserving Sharp Edges in Geometry Images.
Path Differentials for MC Rendering Frank Suykens Department of Computer Science K.U.Leuven, Belgium Dagstuhl 2001: Stochastic methods in Rendering.
Spherical Parameterization and Remeshing Emil Praun, University of Utah Hugues Hoppe, Microsoft Research.
Advanced Computer Graphics (Spring 2005) COMS 4162, Lectures 18, 19: Monte Carlo Integration Ravi Ramamoorthi Acknowledgements.
CSCE 641: Photon Mapping Jinxiang Chai. Outline Rendering equation Photon mapping.
Photon Tracing with Arbitrary Materials Patrick Yau.
Advanced Computer Graphics (Fall 2009) CS 294, Rendering Lecture 5: Monte Carlo Path Tracing Ravi Ramamoorthi
Advanced Computer Graphics (Fall 2010) CS 283, Lecture 10: Global Illumination Ravi Ramamoorthi Some images courtesy.
Advanced Computer Graphics (Spring 2006) COMS 4162, Lecture 20: Monte Carlo Path Tracing Ravi Ramamoorthi Acknowledgements.
Paper by Alexander Keller
Real-Time Diffuse Global Illumination Using Radiance Hints HPG2011 Georgios (George) Papaioannou Dept. of Computer Science Athens University of Economics.
02/11/05© 2005 University of Wisconsin Last Time Direct Lighting Light Transport Equation (LTE) Intro.
Selected Topics in Global Illumination Computation Jaroslav Křivánek Charles University, Prague
-Global Illumination Techniques
The Natural-Constraint Representation of the Path Space for Efficient Light Transport Simulation Anton S. Kaplanyan 1,2 and Johannes Hanika 1 and Carsten.
Monte Carlo I Previous lecture Analytical illumination formula This lecture Numerical evaluation of illumination Review random variables and probability.
Computer graphics & visualization Photon Mapping.
Computer Graphics Global Illumination: Photon Mapping, Participating Media Lecture 12 Taku Komura.
Path Integral Formulation of Light Transport
Photo-realistic Rendering and Global Illumination in Computer Graphics Spring 2012 Hybrid Algorithms K. H. Ko School of Mechatronics Gwangju Institute.
Using Interactive Ray Tracing for Interactive Global Illumination Computer Graphics Lab Saarland University, Germany
Global Illumination: Radiosity, Photon Mapping & Path Tracing Rama Hoetzlein, 2009 Lecture Notes Cornell University.
- Laboratoire d'InfoRmatique en Image et Systèmes d'information
Real-Time High Quality Rendering
On robust Monte Carlo algorithms for multi-pass global illumination Frank Suykens – De Laet 17 September 2002.
Monte-Carlo Ray Tracing and
PATH INTEGRAL METHODS FOR LIGHT TRANSPORT SIMULATION THEORY & PRACTICE Jaroslav Křivánek Charles University Prague Juan Cañada Next Limit Technologies.
Photo-realistic Rendering and Global Illumination in Computer Graphics Spring 2012 Hybrid Algorithms K. H. Ko School of Mechatronics Gwangju Institute.
Pure Path Tracing: the Good and the Bad Path tracing concentrates on important paths only –Those that hit the eye –Those from bright emitters/reflectors.
02/12/03© 2003 University of Wisconsin Last Time Intro to Monte-Carlo methods Probability.
Computer Graphics III – Monte Carlo integration Direct illumination
Bidirectional Path Sampling Techniques
Thank you for the introduction
Computer Graphics III Winter Term 2015 Organization Jaroslav Křivánek, MFF UK
Slide 1Lastra, 2/14/2016 Monte-Carlo Methods. Slide 2Lastra, 2/14/2016 Topics Kajiya’s paper –Showed that existing rendering methods are approximations.
Monte Carlo Path Tracing
University of Texas at Austin CS395T - Advanced Image Synthesis Spring 2007 Don Fussell Photon Mapping and Irradiance Caching.
Fast Global Illumination Including Specular Effects Xavier Granier 1 George Drettakis 1 Bruce J. Walter 2 1 iMAGIS -GRAVIR/IMAG-INRIA iMAGIS is a joint.
Global Illumination (3) Photon Mapping (1). Overview Light Transport Notation Path Tracing Photon Mapping –Photon Tracing –The Photon Map.
Global Illumination (3) Path Tracing. Overview Light Transport Notation Path Tracing Photon Mapping.
02/9/05© 2005 University of Wisconsin Last Time Lights Next assignment – Implement Kubelka-Munk as a BSDF.
PATH INTEGRAL FORMULATION OF LIGHT TRANSPORT Jaroslav Křivánek Charles University in Prague
Computer graphics III – Multiple Importance Sampling Jaroslav Křivánek, MFF UK
Computer Graphics III Winter Term 2017 Organization
Sampling and Reconstruction of Visual Appearance
Sampling and Reconstruction of Visual Appearance
Recent Advances in Light Transport Simulation Theory & Practice
Ray Tracing via Markov Chain Monte-Carlo Method
Using Flow Textures to Visualize Unsteady Vector Fields
The Rendering Equation
© 2005 University of Wisconsin
Primary Sample Space Path Guiding
Progressive Photon Mapping
Advanced Light Transport in the VFX/Archiviz industry
Virtual Spherical Lights for Many-Light Rendering of Glossy Scenes
Progressive Photon Mapping Toshiya Hachisuka Henrik Wann Jensen
Tips and Tricks for Gradient-domain Rendering Implementation
Bidirectional Path Sampling Techniques
Smoother Subsurface Scattering
(Optimizing) Realistic Rendering with Many-Light Methods
Computer Graphics III Winter Term 2016 Organization
Monte Carlo Path Tracing and Caching Illumination
Path Space Regularization for Robust and Holistic Light Transport
Photon Density Estimation using Multiple Importance Sampling
Presentation transcript:

Anton S. Kaplanyan Karlsruhe Institute of Technology, Germany Path Space Regularization Framework

2 Motivation Why Photon Mapping / Vertex Merging is useful? Caustics/reflected caustics Helps sampling difficult transport paths Why is it not the ultimate technique? Slower convergence on diffuse High memory and b/w requirements Cache efficiency is unpredictable Can we have all pros and no cons? Handling difficult illumination No memory pressure Converges as fast as possible

3 Photon Mapping Cache (photon map) Merging (kernel estimation) Not discussed In this talk Selective, where required

4 Motivation

5

6

7 Non-Sampleable Paths What if all edges of the path contain singularities? No place to perform a connection Impossible to sample with any unbiased sampling [Veach97] Example: Reflected caustics from point light (LSDSE)

8 Excursus: Undecidability of Ray Tracing Pure specular path tracing is undecidable [Reif et al. 1994] Task: Find all specular paths from one fixed point to another (DS + D) Impossible to find all such paths on a Turing machine Given arbitrary specular geometry

9 Non-Sampleable Paths, Part II What if all edges of the path contain singularities? No place to perform connection Impossible to sample with any unbiased sampling Example: Reflected caustics from point light (LSDSE) Reflected caustics: Two undecidable subpaths Such paths cannot be found precisely

10 Photon Mapping (a.k.a. Vertex Merging)

11 Photon Mapping: Under the Hood Photon mapping essentially regularizes specular interactions Regularization angle depends on the connection distance On-surface radius is fixed for a given point It is a known mathematical procedure…

12 Regularization with Mollification

13 Selective Regularization of Path Space Goal: Minimize bias. Why smoothing a regular path? Photon mapping causes uniform bias everywhere Regularize only non-sampleable paths! Only if all edges join a singularity Can only be detected once all interactions are known

14 Regularization Bandwidth Fix on-surface radius, as in photon mapping Convenient to control perceived bias Shoot a reflection ray to detect the distance Optimal bandwidth depends on light field around the path Requeres bias estimation Optimal bandwidth might depend on the goal / application E.g., regularize to sample all image features within given time

15 Reference Comparison of Different Methods

16 Path Tracing Comparison of Different Methods Regular paths

17 Comparison of Different Methods Bidirectional Path Tracing Regular paths

Noisy without cache?

19 Comparison of Different Methods Metropolis Light Transport Regular paths

20 Observations Ordinary MC methods (PT, BDPT, …) need efficient caching Markov chain Monte Carlo (MLT, …) resolves poor caching problem Inherent caching with the current path of Markov chain Regularization is simple to integrate into existing renderer

21 Observations Ordinary MC methods (PT, BDPT, …) need efficient caching Markov chain Monte Carlo (MLT, …) resolves poor caching problem Inherent caching with the current path of Markov chain Regularization is simple to integrate into existing renderer Path Tracer Path Tracer with Regularization

22 Consistency

23 Practical Convergence of MLT Powerful tool to temper the integrand “Improving Robustness of Monte-Carlo Global Illumination with Directional Regularization” brief today by G. Bouchard et al. Can achieve more uniform convergence with MLT MLT + Regularization, 1 minMLT, 1 minReference

24 Practical Convergence of MLT

25 Combination with Manifold Exploration Enables unbiased sampling of non-sampleable paths! Regularization provides local parameterization Manifold exploration explores the surrounding paths Avoids the undecidability By finding a local parameterization Undecidable paths are sampled “almost surely” Unbiased reflected caustics MLT+ME MLT+ME with Regularization

26 Conclusion Vertex merging = regularization in original directional domain Biased Less efficient without cache (photon map) with MC methods Easy to apply to any integration method Finer control over bias Useful for MLT Seed with regularized paths, find exact non-sampleable paths Bootstrap Markov chain mixing for practical rendering Future work How to select the regularization bandwidth? Where and when to regularize arbitrary interactions? Predictive and preview rendering

Thank you for your attention