1Advanced Computer Graphics Advanced Computer Graphics (Fall 2009)(Fall 2009)CS 294, Rendering Lecture 6: Recent Advances in Monte Carlo Offline RenderingRavi Ramamoorthihttp://inst.eecs.berkeley.edu/~cs294-13/fa09Some slides/ideas courtesy Pat Hanrahan, Henrik Jensen DiscussionDiscussion Problems different over years. Initially, how to make rendering a single picture fast. Now, multidimensional effects, multiple images. Include image-based lighting, reflectance. Monte Carlo itself is a well known numerical method. But, many recent insights, more to come Lecture surveys much work in last decade at high-level. Need to read papers for more depth.History and OutlineHistory and Outline Is Monte Carlo Rendering solved? Can it be made more efficient (90s): Irradiance caching takes advantage of coherence Correct sampling: Stratified, Multiple Importance, Bidirectional path tracing, Metropolis… Photon Mapping Work with Image-Based Appearance (02-06) Importance sampling environment maps, BRDFs Multidimensional effects (depth-of field, soft shadows, motion blur) Adaptive multidimensional sampling Cut-based hierarchical integration Frequency space analysis Smoothness of Indirect LightingSmoothness of Indirect LightingDirectIndirectDirect + IndirectIrradiance CachingIrradiance Caching Empirically, (diffuse) interreflections low frequency Therefore, should be able to sample sparsely Irradiance caching samples irradiance at few points on surfaces, and then interpolates Ward, Rubinstein, Clear. SIGGRAPH 88, A ray tracing solution for diffuse interreflectionIrradiance CalculationIrradiance Calculation(, )iLxw() (,)cosiEx L x dwqw=òpositionrotationDerivation in Ward paper()()()()iiiiiiwxE xExwx=åå1()()wxxe=2Algorithm OutlineAlgorithm Outline Find all samples with w(x) > q if ( samples found ) interpolate else compute new irradiance N.B. Subsample the image first and then fill inIrradiance Caching ExampleIrradiance Caching ExampleFinal ImageSample LocationsHistory and OutlineHistory and Outline Is Monte Carlo Rendering solved? Can it be made more efficient (90s): Irradiance caching takes advantage of coherence Correct sampling: Stratified, Multiple Importance, Bidirectional path tracing, Metropolis… Photon Mapping Work with Image-Based Appearance (02-06) Importance sampling environment maps, BRDFs Multidimensional effects (depth-of field, soft shadows, motion blur) Adaptive multidimensional sampling Cut-based hierarchical integration Frequency space analysis Better SamplingBetter Sampling Smarter ways to Monte Carlo sample Long history: Stratified, Importance, Bi-Directional, Multiple Importance, Metropolis Good reference is Veach thesis We only briefly discuss a couple of strategiesD. Mitchell 95, Consequences of stratified sampling in graphicsSpectrally Optimal SamplingSpectrally Optimal SamplingMitchell 913History and OutlineHistory and Outline Is Monte Carlo Rendering solved? Can it be made more efficient (90s): Irradiance caching takes advantage of coherence Correct sampling: Stratified, Multiple Importance, Bidirectional path tracing, Metropolis… Photon Mapping Work with Image-Based Appearance (02-06) Importance sampling environment maps, BRDFs Multidimensional effects (depth-of field, soft shadows, motion blur) Adaptive multidimensional sampling Cut-based hierarchical integration Frequency space analysis Light Ray TracingLight Ray TracingBackwards Ray Tracing[Arvo 86] Step 1. Choose a light ray Step 2. Find ray-surface intersection Step 3. Reflect or transmitu = Uniform()if u < reflectance(x)Choose new direction d ~ BRDF(O|I)goto Step 2 else if u < reflectance(x)+transmittance(x)Choose new direction d ~ BTDF(O|I)goto Step 2 else // absorption=1–reflectance-transmittanceterminate on surface; deposit energyPath Tracing: From LightsBidirectional Path TracingBidirectional Path TracingPath pyramid (k = l + e = total number of bounces)ComparisonComparison4Why Photon Map?Why Photon Map? Some visual effects like caustics hard with standard path tracing from eye May usually miss light source altogether Instead, store “photons” from light in kd-tree Look-up into this as needed Combines tracing from light source, and eye Similar to bidirectional path tracing, but compute photon map only once for all eye rays Global Illumination using Photon Maps H. Jensen. Rendering Techniques (EGSR 1996), pp 21-30. (Also book: Realistic Image Synthesis using Photon Mapping)CausticsCausticsSlides courtesy Henrik Wann JensenPath Tracing: 1000 paths/pixelNote noise in causticsCausticsCausticsPhoton Mapping: 10000 photons50 photons in radiance estimateReflections Inside a Metal RingReflections Inside a Metal Ring50000 photons 50 photons to estimate radianceCaustics on Glossy SurfacesCaustics on Glossy Surfaces340000 photons, 100 photons in radiance estimate340000 photons, 100 photons in radiance estimateHDR Environment IlluminationHDR Environment Illumination5Global IlluminationGlobal IlluminationDirect IlluminationDirect IlluminationSpecularSpecularReflectionReflectionCausticsCausticsIndirect IlluminationIndirect IlluminationCornell BoxCornell Box200000 global photons, 50000 caustic photons6Box: Global PhotonsBox: Global PhotonsMiesMiesHouse: Swimming PoolHouse: Swimming PoolHistory and OutlineHistory and Outline Is Monte Carlo Rendering solved? Can it be made more efficient (90s): Irradiance caching takes advantage of coherence Correct sampling: Stratified, Multiple Importance, Bidirectional path tracing, Metropolis… Photon Mapping Work with Image-Based Appearance (02-06) Importance sampling environment maps, BRDFs Multidimensional effects (depth-of field, soft shadows, motion blur) Adaptive multidimensional sampling Cut-based hierarchical integration Frequency space analysis ImageImage--Based AppearanceBased Appearance Standard global illumination is difficult, but the emitters and reflective properties are simple In mid-1990s, interest in appearance acquired from real world, such as image-based lighting Environment Maps, measured BRDFs. These are functions. E.g. any of million pixels emitter How to (importance) sample lighting, BRDFs? Agarwal et al. SIGGRAPH 03, Lawrence et al. SIGGRAPH 04, Clarberg et al. SIGGRAPH 05Structured Importance
View Full Document
Unlocking...