We introduce NeuralVDB, which improves on an existing industry standard for efficient storage of sparse volumetric data, denoted VDB [Museth 2013 ], by leveraging recent advancements in machine learning. Our novel hybrid data structure can reduce the memory footprints of VDB volumes by orders of magnitude, while maintaining its flexibility and only incurring small (user-controlled) compression errors. Specifically, NeuralVDB replaces the lower nodes of a shallow and wide VDB tree structure with multiple hierarchical neural networks that separately encode topology and value information by means of neural classifiers and regressors respectively. This approach is proven to maximize the compression ratio while maintaining the spatial adaptivity offered by the higher-level VDB data structure. For sparse signed distance fields and density volumes, we have observed compression ratios on the order of 10× to more than 100× from already compressed VDB inputs, with little to no visual artifacts. Furthermore, NeuralVDB is shown to offer more effective compression performance compared to other neural representations such as Neural Geometric Level of Detail [Takikawa et al. 2021 ], Variable Bitrate Neural Fields [Takikawa et al. 2022a ], and Instant Neural Graphics Primitives [Müller et al. 2022 ]. Finally, we demonstrate how warm-starting from previous frames can accelerate training, i.e., compression, of animated volumes as well as improve temporal coherency of model inference, i.e., decompression.

We present a neural network-based simulation super-resolution framework that can efficiently and realistically enhance a facial performance produced by a low-cost, real-time physics-based simulation to a level of detail that closely approximates that of a reference-quality off-line simulator with much higher resolution (27× element count in our examples) and accurate physical modeling. Our approach is rooted in our ability to construct a training set of paired frames, from the low- and high-resolution simulators respectively, that are in semantic correspondence with each other. We use face animation as an exemplar of such a simulation domain, where creating this semantic congruence is achieved by simply dialing in the same muscle actuation controls and skeletal pose in the two simulators. Our proposed neural network super-resolution framework generalizes from this training set to unseen expressions, compensates for modeling discrepancies between the two simulations due to limited resolution or cost-cutting approximations in the real-time variant, and does not require any semantic descriptors or parameters to be provided as input, other than the result of the real-time simulation. We evaluate the efficacy of our pipeline on a variety of expressive performances and provide comparisons and ablation experiments for plausible variations and alternatives to our proposed scheme. Our code is available at https://github.com/hjoonpark/3d-sim-super- res.git.

The central argument against data-driven methods in computer graphics rests on the curse of dimensionality: it is intractable to precompute "everything" about a complex space. In this paper, we challenge that assumption by using several thousand CPU-hours to perform a massive exploration of the space of secondary clothing effects on a character animated through a large motion graph. Our system continually explores the phase space of cloth dynamics, incrementally constructing a secondary cloth motion graph that captures the dynamics of the system. We find that it is possible to sample the dynamical space to a low visual error tolerance and that secondary motion graphs containing tens of gigabytes of raw mesh data can be compressed down to only tens of megabytes. These results allow us to capture the effect of high-resolution, off-line cloth simulation for a rich space of character motion and deliver it efficiently as part of an interactive application.

The explosive or volcanic scenes in motion pictures involve complex turbulent flow as its temperature and density vary in space. To simulate this turbulent flow of an inhomogeneous fluid, we propose a simple and efficient framework. Instead of explicitly computing the complex motion of this fluid dynamical instability, we first approximate the average motion of the fluid. Then, the high-resolution dynamics is computed using our new extended version of the vortex particle method with baroclinity. This baroclinity term makes turbulent effects by generating new vortex particles according to temperature/density distributions. Using our method, we efficiently simulated a complex scene with varying density and temperature.

In this paper, we propose a simple and efficient framework for simulating dispersed bubble flow. Instead of modeling the complex hydrodynamics of numerous small bubbles explicitly, our method approximates the average motion of these bubbles using a continuum multiphase solver. Then, the subgrid interactions among bubbles are computed using our new stochastic solver. Using the proposed scheme, we can efficiently simulate complex scenes with millions of bubbles.

This paper presents a novel framework for simulating the stretching and wiggling of liquids. We demonstrate that complex phase-interface dynamics can be effectively simulated by introducing the Eulerian vortex sheet method, which focuses on the vorticity at the interface (rather than the whole domain). We extend this model to provide user control for the production of visual effects. Then, the generated fluid flow creates complex surface details, such as thin and wiggling fluid sheets. To capture such high-frequency features efficiently, this work employs a denser grid for surface tracking in addition to the (coarser) simulation grid. In this context, the paper proposes a filter, called the liquid-biased filter, which is able to downsample the surface in the high-resolution grid into the coarse grid without unrealistic volume loss resulting from aliasing error. The proposed method, which runs on a single PC, realistically reproduces complex fluid scenes.

Abstract In this paper, we propose a new constrained interpolation profile (CIP) method that is stable and accurate but requires less amount of computation compared to existing CIP‐based solvers. CIP is a high‐order fluid advection solver that can reproduce rich details of fluids. It has third‐order accuracy but its computation is performed over a compact stencil. These advantageous features of CIP are, however, diluted by the following two shortcomings: (1) CIP contains a defect in the utilization of the grid data, which makes the method suitable only for simulations with a tight CFL restriction; and (2) CIP does not guarantee unconditional stability. There have been several attempts to fix these problems in CIP, but they have been only partially successful. The solutions that fixed both problems ended up introducing other undesirable features, namely increased computation time and/or reduced accuracy. This paper proposes a novel modification of the original CIP method that fixes all of the above problems without increasing the computational load or reducing the accuracy. Both quantitative and visual experiments were performed to test the performance of the new CIP in comparison to existing fluid solvers. The results show that the proposed method brings significant improvements in both accuracy and speed.

We present a new fluid simulation technique that significantly reduces the nonphysical dissipation of velocity. The proposed method is based on an apt use of particles and derivative information. We note that a major source of numerical dissipation in the conventional Navier-Stokes equations solver lies in the advection step. Hence, starting with the conventional grid-based simulator, when the details of fluid movements need to be simulated, we replace the advection part with a particle simulator. When swapping between the grid-based and particle-based simulators, the physical quantities such as the level set and velocity must be converted. For this purpose, we develop a novel dissipation-suppressing conversion procedure that utilizes the derivative information stored in the particles, as well as in the grid points. For the fluid regions where such details are not needed, the advection is simulated using an octree-based constrained interpolation profile (CIP) solver, which we develop in this work. Through several experiments, we show that the proposed technique can reproduce the detailed movements of high-Reynolds-number fluids such as droplets/bubbles, thin water sheets, and whirlpools. The increased accuracy in the advection, which forms the basis of the proposed technique, can also be used to produce better results in larger scale fluid simulations.