The fellowship of Metalog BSDFs

Abstract

The lack of a universal material representation remains a significant challenge in appearance modeling, limiting the standardization of Bidirectional Scattering Distribution Functions (BSDFs) [Guarnera, 2016].In this thesis, we unify various material representations under the flexibility of metalog distributions [Keelin, 2016].To cope with challenging multimodal data commonly encountered in appearance modeling, we introduce metalog mixture models. We obtain the multivariate representation via marginal-conditional metalogs and show that interpolation leads to a feasible continuous representation. Alongside their flexibility in representing various families of distributions, metalogs offer intuitive parameters, exceptional compression, and differentiability. Furthermore, their ease of use and closed-form analytical formulation make them a natural choice for simulation applications. Given these properties, we introduce Metalog BSDFs as a unified representation for appearance modeling. We demonstrate that Metalog BSDFs accurately reproduce both established analytic models and data-driven BSDFs, effectively establishing the Fellowship of Metalog BSDFs.