Physically based shading in theory and practice

At SIGGRAPH 2016, we presented a new real-time area lighting technique for polygonal sources. In this talk, we will show how the underlying framework, based on Linearly Transformed Cosines (LTCs), can be extended to support line and disk lights. We will discuss the theory behind these approaches as well as practical implementation tips and tricks concerning numerical precision and performance.

PDI/DreamWorks was one of the first animation studios to adopt global illumination in production rendering. Concurrently, we have also been developing and applying physically based shading principles to improve the consistency and realism of our material models, while balancing the need for intuitive artistic control required for feature animations.

In this talk, we will start by presenting the evolution of physically based shading in our films. Then we will present some fundamental principles with respect to importance sampling and energy conservation in our BSDF framework with a pragmatic and efficient approach to transimssion fresnel modeling. Finally, we will present our new set of physically plausible production shaders for our new path tracer, which includes our new hard surface shader, our approach to material layering and some new developments in fabric and glitter shading.

Recent advances in shading have led to the use of free-path sampling to better solve complex light transport within volumetric materials. In this talk, we describe how we have implemented these ideas and techniques within a production environment, their application on recent shows---such as Guardians of the Galaxy Vol. 2 and Alien: Covenant---and the effect this has had on artists' workflow within our studio.

This talk presents a practical approach to multilayer, physically based surface rendering, specifically optimized for Forward+ rendering pipelines. The presented pipeline allows for the creation of complex surface by decomposing them into different mediums, each represented by a simple BRDF/BSSRDF and set of simple, physical macro properties, such as thickness, scattering and absorption. The described model is explained via practical examples of common multilayer materials such as car paint, lacquered wood, ice, and semi-translucent plastics. Finally, the talk describes intrinsic implementation details for achieving a low performance budget for 60 Hz titles as well as supporting multiple rendering modes: opaque, alpha blend, and refractive blend.

Pixar's Foundation Materials, PxrSurface and PxrMarschnerHair, began shipping with RenderMan 21.

PxrSurface is the standard surface shader developed in the studio for Finding Dory and used more recently for Cars 3 and Coco. This shader contains nine lobes that cover the entire gamut of surface materials for these two films: diffuse, three specular, iridescence, fuzz, subsurface, single scatter and a glass lobe. Each of these BxDF lobes is energy conserving, but conservation is not enforced between lobes on the surface level. We use parameter layering methods to feed a PxrSurface with pre-layered material descriptions. This simultaneously allows us the flexibility of a multilayered shading pipeline together with efficient and consistent rendering behavior.

We also implemented our individual BxDFs with the latest state-of-the-art techniques. For example, our three specular lobes can be switched between Beckmann and GGX modes. Many compound materials have multiple layers of specular; these lobes interact with each other modulated by the Fresnel effect of the clearcoat layer. We also leverage LEADR mapping to recreate sub-displacement micro features such as metal flakes and clearcoat scratches.

Another example is that PxrSurface ships with Jensen, d'Eon and Burley diffusion profiles. Additionally, we implemented a novel subsurface model using path-traced volumetric scattering, which represents a significant advancement. It captures zero and single scattering events of subsurface scattering implicit to the path-tracing algorithm. The user can adjust the phase-function of the scattering events and change the extinction profiles, and it also comes with standardized color inversion features for intuitive albedo input. To the best of our knowledge, this is the first commercially available rendering system to model these features and the rendering cost is comparable to classic diffusion subsurface scattering models.

PxrMarschnerHair implements Marschner's seminal hair illumination model with importance sampling. We also account for the residual energy left after the R, TT, TRT and glint lobes, through a fifth diffuse lobe. We show that this hair surface shader can reproduce dark and blonde hair effectively in a path-traced production context. Volumetric scattering from fiber to fiber changes the perceived hue and saturation of a groom, so we also provide a color inversion scheme to invert input albedos, such that the artistic inputs are straightforward and intuitive.

Two years ago, the rendering and shading groups at Sony Imageworks embarked on a project to review the structure of our physically based shaders in an effort to simplify their implementation, improve quality and pave the way to take advantage of future improvements in light transport algorithms.

We started from classic microfacet BRDF building blocks and investigated energy conservation and artist friendly parametrizations. We continued by unifying volume rendering and subsurface scattering algorithms and put in place a system for medium tracking to improve the setup of nested media. Finally, from all these building blocks, we rebuilt our artist-facing shaders with a simplified interface and a more flexible layering approach through parameter blending.

Our talk will discuss the details of our various building blocks, what worked and what didn't, as well as some future research directions we are still interested in exploring.