研究目的
To efficiently render alias-free ground truth-quality images of scenes with microfacet BRDFs, high-frequency normal variation, all-frequency lighting, both with and without global illumination, addressing the high computational cost and memory requirements of existing methods.
研究成果
The proposed scalable appearance model significantly outperforms state-of-the-art methods in speed (up to 30× faster) and memory efficiency (15% footprint) for rendering high-frequency microfacet materials under complex lighting and global illumination. It achieves alias-free, ground truth-quality images with modest memory requirements and is adaptable to various rendering scenarios.
研究不足
The method assumes axis-aligned footprints in texture-space, locally flat macro-geometry, and no view/light variation within the footprint. It is limited to geometric optics and does not incorporate wave optics. The histogram resolution and SH order choices may affect accuracy, and the approach may not handle arbitrary convex footprints efficiently without extensions.
1：Experimental Design and Method Selection:
The paper presents a filtered appearance model that combines a spherical histogram for multi-scale normal distribution functions (NDFs) with a spherical harmonics (SH) basis for efficient integration. It includes a novel half-vector histogram scheme and adaptive basis-space integration to handle complex lighting and global illumination.
2：Sample Selection and Data Sources:
High-resolution normal maps (e.g., 2048x2048) are used as input for explicit normal variation. Scenes include objects like cutlery, snails, torus, and kettle with environmental and point lighting.
3：List of Experimental Equipment and Materials:
A dual Intel Xeon E5-2683 with 32 cores and 128 GB of RAM is used for rendering. Normal maps and environment maps are specified, but no specific brands or models are mentioned.
4：Experimental Procedures and Operational Workflow:
The method involves initializing a spherical histogram SAT, precomputing SH coefficients for Gaussian roughness lobes and lighting, and performing double-product integrals in SH space. It is integrated into a path tracer for direct and global illumination, with comparisons to state-of-the-art methods like Yan et al. 2016 and Belcour et al.
5：Data Analysis Methods:
20 Performance is evaluated based on render times, memory usage, and visual comparison to ground truth. Speed-up factors and memory footprints are quantified.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
