27. connecting analytical and measured brdfs
Name: Tiancheng Sun
Grad Year: 2022
The bidirectional reflectance function (BRDF) is crucial for rendering the appearance of real-world materials. Two kinds of models are usually used to express the BRDF: Analytical models give a parametric approximation of the BRDF, and are easier to manipulate and render. Measured or data-driven models capture reflectance values of real-world objects; they are more accu- rate by definition, but can be harder to work with, and less flexible to edit or manipulate. In this work, we take important steps towards understanding and bridging the gap between analytical and measured BRDFs. First, we develop a robust method for separating a measured BRDF into diffuse and specular components, as is commonly done in analytical models but has been difficult previously to do explicitly for data-driven reflectance. This enables edits such as changing the relative amounts of diffuse and specular reflectance and the color of each component, which is trivial for analytic BRDFs but was not easy to do for earlier data-driven BRDF models. Our diffuse-specular separation also enables an analysis of each of these compo- nents separately in the MERL dataset, and we demonstrate a more intuitive and lower-dimensional PCA models than Nielsen et al. . Next, we analyze the relationship between a particular analytic BRDF model, the GGX model, and the specular component of measured BRDFs represented in a 3D log-PC space. We show that analytical BRDFs lie in a manifold of the low-dimensional space of measured BRDFs, which provides a new con- nection between analytic and data-driven reflectance. Finally, we construct a simple neural network-based mapping between analytic parameters and low-dimensional measured BRDFs, enabling direct translation between them, without expensive optimization or fitting. We also demonstrate applications such as editing measured BRDFs through their analytic parameters, and fitting analytic BRDFs from optimal sparse directional measurements.
Industry Application Area(s)