Many materials, such as sand, rice, wheat, or other kinds of seeds, consist of numerous individual grains that determine the visual appearance of these materials. When generating images of these mixtures, the primary challenge is to simulate the interaction of light with each individual grain. While subsurface scattering effects are crucial for producing realistic images, the computation of light transport using standard path tracing methods for each grain can be prohibitively expensive. Although there have been several methods developed to address this issue, they all assume that bounding spheres of individual grains do not intersect. This restriction limits the application of these methods to almost spherical grains. Nonetheless, various grains, such as seeds and rice, are non-spherical, making this assumption lead to impractical stackings in situations involving coarse-grained materials. We address this issue by presenting a subsurface scattering model that utilizes a neural network and is trained using an optimal transport framework. Our model surpasses path tracing approaches conclusively, allowing for efficient rendering of granular mixtures that were previously unfeasible. Additionally, this method can be utilized in large-scale procedural generated scenes based on sphere packings and obtains similar results as previous methods in these cases.

 

Citation:

D. Zingsheim and R. Klein, “Learning subsurface scattering solutions of tightly-packed granular media using optimal transport,” Computers & Graphics, vol. 119, p. 103895, Feb. 2024, doi: 10.1016/j.cag.2024.103895.

 

 

More Information:

Open source: https://doi.org/10.1016/j.cag.2024.103895