Neural networks have proven effective at solving difficult problems but designing their architectures can be challenging, even for image classification problems alone. Our goal is to minimize human participation, so we employ evolutionary algorithms to discover such networks automatically. Despite significant computational requirements, we show that it is now possible to evolve models with accuracies within the range of those published in the last year. Specifically, we employ simple evolutionary techniques at unprecedented scales to discover models for the CIFAR-10 and CIFAR-100 datasets, starting from trivial initial conditions and reaching accuracies of 94.6\% (95.6\% for ensemble) and 77.0\%, respectively. To do this, we use novel and intuitive mutation operators that navigate large search spaces; we stress that no human participation is required once evolution starts and that the output is a fully-trained model. Throughout this work, we place special emphasis on the repeatability of results, the variability in the outcomes and the computational requirements.

We present a novel approach to guiding physically based particle simulations using boundary conditions. Unlike commonly used ad hoc particle techniques for adding and removing the material from a simulation, our approach is principled by utilizing the concept of volumetric flux. Artists are provided with a simple yet powerful primitive called a fluxed animated boundary (FAB) , allowing them to specify a control shape and a material flow field. The system takes care of enforcing the corresponding boundary conditions and necessary particle reseeding. We show how FABs can be used artistically or physically. Finally, we demonstrate production examples that show the efficacy of our method.

Hybrid Lagrangian/Eulerian simulation is commonplace in computer graphics for fluids and other materials undergoing large deformation. In these methods, particles are used to resolve transport and topological change, while a background Eulerian grid is used for computing mechanical forces and collision responses. Particle-in-Cell (PIC) techniques, particularly the Fluid Implicit Particle (FLIP) variants have become the norm in computer graphics calculations. While these approaches have proven very powerful, they do suffer from some well known limitations. The original PIC is stable, but highly dissipative, while FLIP, designed to remove this dissipation, is more noisy and at times, unstable. We present a novel technique designed to retain the stability of the original PIC, without suffering from the noise and instability of FLIP. Our primary observation is that the dissipation in the original PIC results from a loss of information when transferring between grid and particle representations. We prevent this loss of information by augmenting each particle with a locally affine, rather than locally constant, description of the velocity. We show that this not only stably removes the dissipation of PIC, but that it also allows for exact conservation of angular momentum across the transfers between particles and grid.

Evaluating transmittance within participating media is a fundamental operation required by many light transport algorithms. We present ratio tracking and residual tracking , two complementary techniques that can be combined into an efficient, unbiased estimator for evaluating transmittance in complex heterogeneous media. In comparison to current approaches, our new estimator is unbiased, yields high efficiency, gracefully handles media with wavelength dependent extinction, and bridges the gap between closed form solutions and purely numerical, unbiased approaches. A key feature of ratio tracking is its ability to handle negative densities. This in turn enables us to separate the main part of the transmittance function, handle it analytically, and numerically estimate only the residual transmittance. In addition to proving the unbiasedness of our estimators, we perform an extensive empirical analysis to reveal parameters that lead to high efficiency. Finally, we describe how to integrate the new techniques into a production path tracer and demonstrate their benefits over traditional unbiased estimators.

In this paper, we introduce a novel material point method for heat transport, melting and solidifying materials. This brings a wider range of material behaviors into reach of the already versatile material point method. This is in contrast to best-of-breed fluid, solid or rigid body solvers that are difficult to adapt to a wide range of materials. Extending the material point method requires several contributions. We introduce a dilational/deviatoric splitting of the constitutive model and show that an implicit treatment of the Eulerian evolution of the dilational part can be used to simulate arbitrarily incompressible materials. Furthermore, we show that this treatment reduces to a parabolic equation for moderate compressibility and an elliptic, Chorin-style projection at the incompressible limit. Since projections are naturally done on marker and cell (MAC) grids, we devise a staggered grid MPM method. Lastly, to generate varying material parameters, we adapt a heat-equation solver to a material point framework.

Abstract Ray‐traced global illumination (GI) is becoming widespread in production rendering but incoherent secondary ray traversal limits practical rendering to scenes that fit in memory. Incoherent shading also leads to intractable performance with production‐scale textures forcing renderers to resort to caching of irradiance, radiosity, and other values to amortize expensive shading. Unfortunately, such caching strategies complicate artist workflow, are difficult to parallelize effectively, and contend for precious memory. Worse, these caches involve approximations that compromise quality. In this paper, we introduce a novel path‐tracing framework that avoids these tradeoffs. We sort large, potentially out‐of‐core ray batches to ensure coherence of ray traversal. We then defer shading of ray hits until we have sorted them, achieving perfectly coherent shading and avoiding the need for shading caches.

Snow is a challenging natural phenomenon to visually simulate. While the graphics community has previously considered accumulation and rendering of snow, animation of snow dynamics has not been fully addressed. Additionally, existing techniques for solids and fluids have difficulty producing convincing snow results. Specifically, wet or dense snow that has both solid- and fluid-like properties is difficult to handle. Consequently, this paper presents a novel snow simulation method utilizing a user-controllable elasto-plastic constitutive model integrated with a hybrid Eulerian/Lagrangian Material Point Method. The method is continuum based and its hybrid nature allows us to use a regular Cartesian grid to automate treatment of self-collision and fracture. It also naturally allows us to derive a grid-based semi-implicit integration scheme that has conditioning independent of the number of Lagrangian particles. We demonstrate the power of our method with a variety of snow phenomena including complex character interactions.

We present a new algorithm for near-interactive simulation of skeleton driven, high resolution elasticity models. Our methodology is used for soft tissue deformation in character animation. The algorithm is based on a novel discretization of corotational elasticity over a hexahedral lattice. Within this framework we enforce positive definiteness of the stiffness matrix to allow efficient quasistatics and dynamics. In addition, we present a multigrid method that converges with very high efficiency. Our design targets performance through parallelism using a fully vectorized and branch-free SVD algorithm as well as a stable one-point quadrature scheme. Since body collisions, self collisions and soft-constraints are necessary for real-world examples, we present a simple framework for enforcing them. The whole approach is demonstrated in an end-to-end production-level character skinning system.

We present a method for generating art-directable volumetric effects, ranging from physically-accurate to non-physical results. Our system mimics the way experienced artists think about volumetric effects by using an intuitive lighting primitive, and decoupling the modeling and shading of this primitive. To accomplish this, we generalize the physically-based photon beams method to allow arbitrarily programmable simulation and shading phases. This provides an intuitive design space for artists to rapidly explore a wide range of physically-based as well as plausible, but exaggerated, volumetric effects. We integrate our approach into a real-world production pipeline and couple our volumetric effects to surface shading.

Hair simulation remains one of the most challenging aspects of creating virtual characters. Most research focuses on handling the massive geometric complexity of hundreds of thousands of interacting hairs. This is accomplished either by using brute force simulation or by reducing degrees of freedom with guide hairs. This paper presents a hybrid Eulerian/Lagrangian approach to handling both self and body collisions with hair efficiently while still maintaining detail. Bulk interactions and hair volume preservation is handled efficiently and effectively with a FLIP based fluid solver while intricate hair-hair interaction is handled with Lagrangian self-collisions. Thus the method has the efficiency of continuum/guide based hair models with the high detail of Lagrangian self-collision approaches.

Turbulent vortices in fluid flows are crucial for a visually interesting appearance. Although there has been a significant amount of work on turbulence in graphics recently, these algorithms rely on the underlying simulation to resolve the flow around objects. We build upon work from classical fluid mechanics to design an algorithm that allows us to accurately precompute the turbulence being generated around an object immersed in a flow. This is made possible by modeling turbulence formation based on an averaged flow field, and relying on universal laws describing the flow near a wall. We precompute the confined vorticity in the boundary layer around an object, and simulate the boundary layer separation during a fluid simulation. Then, a turbulence model is used to identify areas where this separated layer will transition into actual turbulence. We sample these regions with vortex particles, and simulate the further dynamics of the vortices based on these particles. We will show how our method complements previous work on synthetic turbulence, and yields physically plausible results. In addition, we demonstrate that our method can efficiently compute turbulent flows around a variety of objects including cars, whisks, as well as boulders in a river flow. We can even apply our model to precomputed static flow fields, yielding turbulent dynamics without a costly simulation.

In this paper we simulate high resolution cloth consisting of up to 2 million triangles which allows us to achieve highly detailed folds and wrinkles. Since the level of detail is also influenced by object collision and self collision, we propose a more accurate model for cloth-object friction. We also propose a robust history-based repulsion/collision framework where repulsions are treated accurately and efficiently on a per time step basis. Distributed memory parallelism is used for both time evolution and collisions and we specifically address Gauss-Seidel ordering of repulsion/collision response. This algorithm is demonstrated by several high resolution and high-fidelity simulations.

Our goal is to simulate the full hair geometry, consisting of approximately one hundred thousand hairs on a typical human head. This will require scalable methods that can simulate every hair as opposed to only a few guide hairs. Novel to this approach is that the individual hair/hair interactions can be modeled with physical parameters (friction, static attraction, etc.) at the scale of a single hair as opposed to clumped or continuum interactions. In this vein, we first propose a new altitude spring model for preventing collapse in the simulation of volumetric tetrahedra, and we show that it is also applicable both to bending in cloth and torsion in hair. We demonstrate that this new torsion model for hair behaves in a fashion similar to more sophisticated models with significantly reduced computational cost. For added efficiency, we introduce a semi-implicit discretization of standard springs that makes them truly linear in multiple spatial dimensions and thus unconditionally stable without requiring Newton-Raphson iteration. We also simulate complex hair/hair interactions including sticking and clumping behavior, collisions with objects (e.g. head and shoulders) and self-collisions. Notably, in line with our goal to simulate the full head of hair, we do not generate any new hairs at render time.

We explore the emerging application area of physics-based simulation for computer animation and visual special effects. In particular, we examine its parallelization potential and characterize its behavior on a chip multiprocessor (CMP). Applications in this domain model and simulate natural phenomena, and often direct visual components of motion pictures. We study a set of three workloads that exemplify the span and complexity of physical simulation applications used in a production environment: fluid dynamics, facial animation, and cloth simulation. They are computationally demanding, requiring from a few seconds to several minutes to simulate a single frame; therefore, they can benefit greatly from the acceleration possible with large scale CMPs.

The particle level set method has proven successful for the simulation oftwoseparate regions (such as water and air, or fuel and products). In this paper, we propose a novel approach to extend this method to the simulation of as many regions as desired. The various regions can be liquids (or gases) of any type with differing viscosities, densities, viscoelastic properties, etc. We also propose techniques for simulating interactions between materials, whether it be simple surface tension forces or more complex chemical reactions with one material converting to another or two materials combining to form a third. We use a separate particle level set method for each region, and propose a novel projection algorithm that decodes the resulting vector of level set values providing a "dictionary" that translates between them and the standard single-valued level set representation. An additional difficulty occurs since discretization stencils (for interpolation, tracing semi-Lagrangian rays, etc.) cross region boundaries naively combining non-smooth or even discontinuous data. This has recently been addressed via ghost values, e.g. for fire or bubbles. We instead propose a new paradigm that allows one to incorporate physical jump conditions in data "on the fly," which is significantly more efficient for multiple regions especially at triple points or near boundaries with solids.

We present a novel method for solid/fluid coupling that can treat infinitesimally thin solids modeled by a lower dimensional triangulated surface. Since classical solid/fluid coupling algorithms rasterize the solid body onto the fluid grid, an entirely new approach is required to treat thin objects that do not contain an interior region. Robust ray casting is used to augment a number of interpolation, finite difference and rendering techniques so that fluid does not leak through the triangulated surface. Moreover, we propose a technique for properly enforcing incompressibility so that fluid does not incorrectly compress (and appear to lose mass) near the triangulated surface. This allows for the robust interaction of cloth and shells with thin sheets of water. The proposed method works for both rigid body shells and for deformable manifolds such as cloth, and we present a two way coupling technique that allows the fluid's pressure to affect the solid. Examples illustrate that our method performs well, especially in the difficult case of water and cloth where it produces visually rich interactions between the particle level set method for treating the water/air interface and our newly proposed method for treating the solid/fluid interface. We have implemented the method on both uniform and adaptive octree grids.

Vorticity confinement reintroduces the small scale detail lost when using efficient semi-Lagrangian schemes for simulating smoke and fire. However, it only amplifies the existing vorticity, and thus can be insufficient for highly turbulent effects such as explosions or rough water. We introduce a new hybrid technique that makes synergistic use of Lagrangian vortex particle methods and Eulerian grid based methods to overcome the weaknesses of both. Our approach uses vorticity confinement itself to couple these two methods together. We demonstrate that this approach can generate highly turbulent effects unachievable by standard grid based methods, and show applications to smoke, water and explosion simulations.