Upsampling of Lagrangian Particle Simulations

We present a novel post-processing framework to efficiently upsample low-resolution Lagrangian particle simulations. For these simulations, the number of particles has a major impact on how impressive the results are. However, the particle count is also often constrained by time and hardware limitations. To solve this problem, we instead take coarse simulations as input and upsample them, capturing the bulk behavior, while adding further detail. This significantly improves the visual quality of surface areas, which is most evident in water splashes or thin layers of free-falling sand grains. First, high-resolution particles are created based on the initial input positions. To this end, we propose a new sampling strategy based on surface estimation to closely match the shape implied by the input particles. The motion of the coarse simulation is then transferred to the highresolution output via interpolation, while taking into account ballistic behavior and collisions with static or moving boundaries. For challenging dynamic scenarios, we propose an optional position correction step to improve particle distribution quality and volume conservation. Our pipeline is not tuned to a specific material or simulation method, which we demonstrate for both coarse DEM granular simulations, as well as highly dynamic SPH fluid simulations, using millions of upsampled particles.