Efficient Scene Appearance Aggregation for Level-of-Detail Rendering

TL;DR

This work tackles appearance-preserving level-of-detail rendering for large, complex scenes by introducing a far-field Aggregated Bidirectional Scattering Distribution Function ($\hat{f}$) per voxel. The authors derive a closed-form, factorized ABSDF that separates base material, normals, and visibility, enabling efficient evaluation and sampling while preserving intra-voxel and inter-voxel correlations through a truncated ellipsoid primitive and global visibility terms (AIV and ABV). A complete scene aggregation pipeline is proposed, including multi-level sparse precomputation, directional moments, beta-distributed roughness modeling, and CPCA-based visibility compression, enabling scalable memory use and faster rendering compared to prior LoD approaches. Quantitative and qualitative results across diverse scenes demonstrate higher fidelity—especially for glossy and anisotropic highlights—while achieving asymptotic memory savings and rendering speed advantages over state-of-the-art methods. The approach offers a practical path toward scalable, physically-based rendering for richly detailed environments with instancing and complex materials.

Abstract

Creating an appearance-preserving level-of-detail (LoD) representation for arbitrary 3D scenes is a challenging problem. The appearance of a scene is an intricate combination of both geometry and material models, and is further complicated by correlation due to the spatial configuration of scene elements. We present a novel volumetric representation for the aggregated appearance of complex scenes and an efficient pipeline for LoD generation and rendering. The core of our representation is the Aggregated Bidirectional Scattering Distribution Function (ABSDF) that summarizes the far-field appearance of all surfaces inside a voxel. We propose a closed-form factorization of the ABSDF that accounts for spatially varying and orientation-varying material parameters. We tackle the challenge of capturing the correlation existing locally within a voxel and globally across different parts of the scene. Our method faithfully reproduces appearance and achieves higher quality than existing scene filtering methods while being inherently efficient to render. The memory footprint and rendering cost of our representation are independent of the original scene complexity.

Figures (18)

• Figure 1: An overview of our method. We start by voxelizing a scene such that the voxel size matches the given pixel footprint. For each voxel, we model its aggregated appearance by its ABSDF. To preserve the local spatial correlation, we use a truncated ellipsoid primitive that describes the intra-voxel geometric distribution. To preserve the long-range correlation, we record global aggregated visibility. Both lead to accurate voxel accumulation that is order-independent.
• Figure 2: Schematic diagram of our full factorized ABSDF model. The usages of precomputed tables $M_1, ..., M_5$ are highlighted.
• Figure 3: We select voxels from each scene, highlighted in (a), and compare the ground truth ABSDFs (b)/(e) to our factored ABSDFs (c)/(f) and the appearance network fitted results (d)/(g). Each plot contains $8 \times 8$ 2D outgoing slices in the lat-long coordinate system with different incident directions. Our results achieve better accuracy both qualitatively and quantitatively with lower RMSE. We encourage readers to zoom in for better comparison. Exposure is adjusted for clarity.
• Figure 4: When a scene has orientation-varying material parameters (a), our method (c) captures the view dependency and matches the reference (b), while ignoring it leads to incorrect results (d).
• Figure 5: Compared to the reference, LoD with a simple cube primitive results in a bloated silhouette and worse, structured artifacts on the red plane. With the help of the truncated ellipsoid primitive, our method produces a tighter silhouette and more importantly, artifact free results. We encourage readers to zoom in to better identify the checkerboard-like artifact.