A Multiscale Microfacet Model Based on Inverse Bin Mapping

Abstract

2 Chaos Software 3 Chaos Czech a. s.Figure 1: A photograph of wing mirror (left) with pronounced glint from metallic flakes that served as an inspiration for our wing mirror scene (middle). The metallic flakes are modelled with a 2K normal map with flakes sampled from a GTR distribution (GTR gamma = 1.5, GTR alpha = 0.002) [Bur12]. Additionally, the roughness of the flakes is modelled with a Beckmann distribution with Beckmann alpha = 0.005. The flake roughness contributes to the overall appearance, and is a useful parameter for artistic control. To the right we provide three regions from the same scene rendered with different Beckmann flake roughness (0.0025, 0.01, 0.04). Small perturbations of the roughness of the flakes completely change the behaviour of the glints. The rendering of such nearly specular surfaces requires some form of filtering, the effect of which is shown in our accompanying video. All the renderings in this figure were done with our proposed normal map filtering algorithm.

Neural Histogram‐Based Glint Rendering of Surfaces With Spatially Varying Roughness

“…We borrow from their ideas, improving data representation to increase accuracy and achieve a smaller memory footprint and higher performance while adding support for SV roughness. The approach by Atanosov et al [AWKK21] uses a forest of kd‐trees to accelerate histogram lookup. They set the bin size based on the roughness of the surface.…”

“…Unfortunately, normal map filtering is not a linear process and requires special filtering techniques to keep the appearance of the surface across scales. Dedicated data structures [AWKK21, GGN18] or hierarchical representations [YHMR16, DLW*22] are used to model the underlying normal distribution (NDF) inside the pixel footprint. These techniques can be expensive in memory or not support analytical integration over large light sources.…”

“…Yan et al [YHJ*14] and subsequent work perform 2D Gaussian‐weighting across the footprint. Contrary to this, we use equal‐weighting similar to prior work on discrete 𝒫 ‐NDF [JHY*14,GGN18,AWKK21]: …”

Real‐Time Rendering of Glinty Appearances using Distributed Binomial Laws on Anisotropic Grids

“…Chermain et al [CCM20] use a similar paradigm expressed in the microfacet framework, which permits including multiple scattering effects for example [CCM19]. Variants of both methods have been explored to reduce the storage or improve the efficiency [GGN18, CSDD20, WHHY20,CSDD21,AWKK21,DLW*22].…”

Constant-Cost Spatio-Angular Prefiltering of Glinty Appearance Using Tensor Decomposition

“…It focuses on fast performance during rendering, but still suffers from prohibitively expensive storage cost in practice. Concurrent work [Atanasov et al 2021] proposes a normal map filtering approach via inverse binning mapping, throwing microstructures into corresponding directional bins as a preprocess, taking advantage of the fact that directional resolution can be fixed regardless of the sizes of pixel footprints. However, this work is limited to Beckmann function.…”