Position-Normal Manifold for Efficient Glint Rendering on High-Resolution Normal Maps

TL;DR

The paper introduces a position-normal manifold framework to render glinty effects from high-resolution normal maps by recasting P-NDF evaluation as 4D mesh intersections and projecting onto the normal plane, yielding an exact and significantly faster solution than prior convolved or GP-based methods. A mesh-cluster hierarchy accelerates large-footprint queries, and an analytic shadow-masking derivation enables accurate shading for both specular and diffuse components. The approach achieves similar glint realism to prior baselines while delivering substantial speedups (notably for large footprints) and provides a practical path to anti-aliased diffuse reflections through the projected-area formulation. Overall, the method enhances high-frequency glint rendering on normal maps and offers insights for efficient microstructure-aware shading in production renderers, with avenues for future work in clustering optimization, multiple scattering, and wave-optics modeling.

Abstract

Detailed microstructures on specular objects often exhibit intriguing glinty patterns under high-frequency lighting, which is challenging to render using a conventional normal-mapped BRDF. In this paper, we present a manifold-based formulation of the glint normal distribution functions (NDF) that precisely captures the surface normal distributions over queried footprints. The manifold-based formulation transfers the integration for the glint NDF construction to a problem of mesh intersections. Compared to previous works that rely on complex numerical approximations, our integral solution is exact and much simpler to compute, which also allows an easy adaptation of a mesh clustering hierarchy to accelerate the NDF evaluation of large footprints. Our performance and quality analysis shows that our NDF formulation achieves similar glinty appearance compared to the baselines but is an order of magnitude faster. Within this framework, we further present a novel derivation of analytical shadow-masking for normal-mapped diffuse surfaces -- a component that is often ignored in previous works.

Figures (18)

• Figure 1: Our position-normal manifold formulation a) converts the $\mathcal{P}$-NDF integration to finding the manifold projections $\mathbf{u}_i$ followed by accumulating a finite number of $\tfrac{k_\mathbf{r}(\mathbf{u}_i-\mathbf{x})}{\vert\det\mathbf{J}(\mathbf{u}_i)\vert}$. In contrast, b) yan2014rendering’s convolved formulation requires computing a complex integral to reason about the NDF. The left images show toy examples of 1D normal and 1D position, and c) shows the full 4D case.
• Figure 2: Normal map texels are placed on a triangle mesh grid ($\mathbf{u}_0\cdots\mathbf{u}_3$), and barycentric interpolation is used to create the continuous $\mathbf{n}(\mathbf{u})$. The right image shows the zoom-in of the dotted region.
• Figure 3: Comparison of $\mathcal{P}$-NDF evaluation. Our analytical evaluation matches the reference given by the binning approach yan2014rendering. It is also close to yan2014rendering's convolved formulation with small intrinsic roughness ($10^{-4}$ here).
• Figure 4: Acceleration structures used by our method, shown as a 2D toy example. a) a min-max hierarchy records the normal triangles' bounding box for every $2^l\!\times\!2^l$ spatial region (shown as $2^l$ here), which helps prune out the never-intersected triangles. b) a cluster hierarchy simplifies the normal map into coarser grids (clusters) and is partitioned by a cut according to the error criteria; the intersection only needs to be checked against the clusters on the cut.
• Figure 5: Mesh cluster hierarchy successfully uses fewer triangles to represent the normal map (column 1,3). This works for $\mathcal{P}$-NDF evaluations of both small (column 2) and large (column 4) footprint.