Syuhei Sato

Deformation of 2D flow fields using stream functions

Recently, visual simulation of fluids has become an important element in many applications, such as movies and computer games. These fluid animations are usually created by physically-based fluid simulation. However, the simulation often requires very expensive computational cost for creating realistic fluid animations. Therefore, when the user tries to create various fluid animations, he or she must execute fluid simulation repeatedly, which requires a prohibitive computational time. To address this problem, this paper proposes a method for deforming velocity fields of fluids while preserving the divergence-free condition. In this paper, we focus on grid-based 2D fluid simulations. Our system allows the user to interactively create various fluid animations from a single set of velocity fields generated by the fluid simulation. In a preprocess, our method converts the input velocity fields into scalar fields representing the stream functions. At run-time, the user deforms the grid representing the scalar stream functions and the deformed velocity fields are then obtained by applying a curl operator to the deformed scalar stream functions. The velocity fields obtained by this process naturally perseveres the divergence-free condition. For the deformation of the grid, we use a method based on Moving Least Squares. The usefulness of our method is demonstrated by several examples.

A data-driven approach for synthesizing high-resolution animation of fire

We propose a simple and efficient data-driven method for synthesizing high-resolution 3D animations of fire from low-resolution fluid simulations. Our method is based on grid-based fluid simulation. The key concept behind our method is to use a precomputed database of high-resolution velocity fields in order to produce small-scale details that are lost in low-resolution velocity fields. The database is constructed by 2D fluid simulation and no high-resolution 3D fluid simulations need to be executed. At run-time, a low-resolution 3D fluid simulation is executed and the velocity field calculated at each time step is approximated by a linear combination of the precomputed velocity fields. This approximation process produces realistic small-scale detail. Using our method, users can efficiently design animations of fire with low-resolution simulation and our method converts them into high-resolution animations. We examine the ability of our method by applying it to simulations of fire under various situations including moving obstacles.

Incompressibility-preserving deformation for fluid flows using vector potentials

Physically based fluid simulations usually require expensive computation cost for creating realistic animations. We present a technique that allows the user to create various fluid animations from an input fluid animation sequence, without the need for repeatedly performing simulations. Our system allows the user to deform the flow field in order to edit the overall fluid behavior. In order to maintain plausible physical behavior, we ensure the incompressibility to guarantee the mass conservation. We use a vector potential for representing the flow fields to realize such incompressibility-preserving deformations. Our method first computes (time-varying) vector potentials from the input velocity field sequence. Then, the user deforms the vector potential, and the system computes the deformed velocity field by taking the curl operator on the vector potential. The incompressibility is thus obtained by construction. We show various examples to demonstrate the usefulness of our method.

A combining method of fluid animations by interpolating flow fields

The computational cost for creating realistic fluid animations by simulation is generally very expensive. In digital production environment, existing precomputed fluid animations are often reused for different scenes in order to reduce the cost for creating scenes containing fluids. However, applying same animations to different scenes produces unacceptable results, so the animation needs to be edited. In order to do this, we develop a method for synthesizing desired flow fields by combining existing flow fields. Our system allows the user to place existing flow fields at arbitrary positions, and combine them by interpolating the regions between these flow fields, to synthesize a new flow field. The interpolation of the flow fields is realized by solving a minimization problem. Our minimization problem consists of two energy functions for smoothly interpolating the velocities and satisfying the incompressibility. Our method can create the desired incompressible flow fields by reusing existing flow fields.

Generating flow fields variations by modulating amplitude and resizing simulation space

The visual simulation of fluids has become an important element in many applications, such as movies and computer games. In these applications, large-scale fluid scenes, such as fire in a village, are often simulated by repeatedly rendering multiple small-scale fluid flows. In these cases, animators are requested to generate many variations of a small-scale fluid flow. This paper presents a method to help animators meet such requirements. Our method enables the user to generate flow field variations from a single simulated dataset obtained by fluid simulation. The variations are generated in both the frequency and spatial domains. Fluid velocity fields are represented using Laplacian eigenfunctions which ensure that the flow field is always incompressible. In generating the variations in the frequency domain, we modulate the coefficients (amplitudes) of the basis functions. To generate variations in the spatial domain, our system expands or contracts the simulation space, then the flow is calculated by solving a minimization problem subject to the resized velocity field. Using our method, the user can easily create various animations from a single dataset calculated by fluid simulation.

Example-Based Turbulence Style Transfer

Generating realistic fluid simulations remains computationally expensive, and animators can expend enormous effort trying to achieve a desired motion. To reduce such costs, several methods have been developed in which high-resolution turbulence is synthesized as a post process. Since global motion can then be obtained using a fast, low-resolution simulation, less effort is needed to create a realistic animation with the desired behavior. While much research has focused on accelerating the low-resolution simulation, the problem controlling the behavior of the turbulent, high-resolution motion has received little attention. In this paper, we show that style transfer methods from image editing can be adapted to transfer the turbulent style of an existing fluid simulation onto a new one. We do this by extending example-based image synthesis methods to handle velocity fields using a combination of patch-based and optimization-based texture synthesis. This approach allows us to take into account the incompressibility condition, which we have found to be a important factor during synthesis. Using our method, a user can easily and intuitively create high-resolution fluid animations that have a desired turbulent motion.

Editing Fluid Animation Using Flow Interpolation

The computational cost for creating realistic fluid animations by numerical simulation is generally expensive. In digital production environments, existing precomputed fluid animations are often reused for different scenes in order to reduce the cost of creating scenes containing fluids. However, applying the same animation to different scenes often produces unacceptable results, so the animation needs to be edited. In order to help animators with the editing process, we develop a novel method for synthesizing the desired fluid animations by combining existing flow data. Our system allows the user to place flows at desired positions and combine them. We do this by interpolating velocities at the boundaries between the flows. The interpolation is formulated as a minimization problem of an energy function, which is designed to take into account the inviscid, incompressible Navier-Stokes equations. Our method focuses on smoke simulations defined on a uniform grid. We demonstrate the potential of our method by showing a set of examples, including a large-scale sandstorm created from a few flow data simulated in a small-scale space.

Example-based synthesis of turbulence by flow field style transfer

The computational cost of physically-based fluid simulation for generating realistic animations is expensive. Therefore, it takes a long time for an animator to create the desired animation. To reduce such costs, several methods have been developed that employ an approach in which turbulence is synthesized as a post process. Since, with this approach, global motion can be obtained using low-resolution fluid simulation, realistic animations with the desired behavior can be created at low cost. In the field of image editing, style transfer methods, which can efficiently achieve the desired stylization by transferring features of an example image to a user-specified image, have been proposed. We combine these concepts and present a novel style transfer method for turbulence by reusing the existing fluid animations. Our system allows the user to transfer turbulent motion from one flow field to other flow fields. We do this by extending example-based image synthesis methods to the flow field: a patch-based synthesis and an optimization-based texture synthesis. Our method is designed such that the resulting flow fields satisfy the incompressibility condition. Our method can intuitively and easily create high-resolution fluid animations with the desired turbulent motion.

Feedback control of fire simulation based on computational fluid dynamics

Visual simulation of fire plays an important role in many applications, such as movies and computer games. In these applications, artists are often requested to synthesize realistic fire with a particular behavior. To meet such requirement, we present a feedback control method for fire simulations. The user can design the shape of fire by placing a set of control points. Our method generates a force field and automatically adjusts a temperature at a fire source, based on user specified control points. Experimental results show that our method can control the fire shape.

Combining multiple flow fields for editing existing fluid animations

In this paper, we develop a method for synthesizing desired flow fields by combining existing multiple flow fields. Our system allows the user to specify arbitrary regions of the precomputed flow fields and combine them to synthesize a new flow field. In order to maintain plausible physical behavior, we ensure the incompressibility for the combined flow field. To address this, we use stream functions for representing the flow fields. However, there exist discontinuities at the boundaries between the combined flow fields, resulting in unnatural animation of fluids. In order to remove the discontinuities, we apply Poisson image editing to the stream functions.