Abstract
Rough metal seen at a grazing angle, like sunset glare on a wet road or a metal floor under a low camera, is where a path tracer struggles most. Drawing microfacet normals from the GGX distribution wastes many samples on facets the viewer cannot see, and the samples that survive carry weights with no upper bound. This video builds the fix step by step: the half-vector Jacobian, the distribution of visible normals, and the spherical-cap construction of Dupuy and Benyoub that samples it. On a galvanized steel sluice gate at , visible-normal sampling removes the back-facing draws (27.9% of naive draws at ) and lowers the per-sample standard deviation by 3.5× at and 5.7× at . In the ray-traced scene the gain shrinks to 1.6×, and with an 8× sun on the mirror direction it reverses slightly (0.926×), which is the case light sampling with multiple importance sampling is built for.
Key figures





References
- E. Heitz. Sampling the GGX Distribution of Visible Normals. Journal of Computer Graphics Techniques 7(4), 2018.
- J. Dupuy, A. Benyoub. Sampling Visible GGX Normals with Spherical Caps. Computer Graphics Forum 42(8), 2023.
- B. Walter, S. R. Marschner, H. Li, K. E. Torrance. Microfacet Models for Refraction through Rough Surfaces. Eurographics Symposium on Rendering, 2007.