SIGGRAPH 2020 论文列表

We present a method for animating yarn-level cloth effects using a thin-shell solver. We accomplish this through numerical homogenization: we first use a large number of yarn-level simulations to build a model of the potential energy density of the cloth, and then use this energy density function to compute forces in a thin shell simulator. We model several yarn-based materials, including both woven and knitted fabrics. Our model faithfully reproduces expected effects like the stiffness of woven fabrics, and the highly deformable nature and anisotropy of knitted fabrics. Our approach does not require any real-world experiments nor measurements; because the method is based entirely on simulations, it can generate entirely new material models quickly, without the need for testing apparatuses or human intervention. We provide data-driven models of several woven and knitted fabrics, which can be used for efficient simulation with an off-the-shelf cloth solver.

We present a class of display designs combining holographic optics, directional backlighting, laser illumination, and polarization-based optical folding to achieve thin, lightweight, and high performance near-eye displays for virtual reality. Several design alternatives are proposed, compared, and experimentally validated as prototypes. Using only thin, flat films as optical components, we demonstrate VR displays with thicknesses of less than 9 mm, fields of view of over 90° horizontally, and form factors approaching sunglasses. In a benchtop form factor, we also demonstrate a full color display using wavelength-multiplexed holographic lenses that uses laser illumination to provide a large gamut and highly saturated color. We show experimentally that our designs support resolutions expected of modern VR headsets and can scale to human visual acuity limits. Current limitations are identified, and we discuss challenges to obtain full practicality.

We propose Hierarchical Optimization Time Integration (HOT) for efficient implicit timestepping of the material point method (MPM) irrespective of simulated materials and conditions. HOT is an MPM-specialized hierarchical optimization algorithm that solves nonlinear timestep problems for large-scale MPM systems near the CFL limit. HOT provides convergent simulations out of the box across widely varying materials and computational resolutions without parameter tuning. As an implicit MPM timestepper accelerated by a custom-designed Galerkin multigrid wrapped in a quasi-Newton solver, HOT is both highly parallelizable and robustly convergent. As we show in our analysis, HOT maintains consistent and efficient performance even as we grow stiffness, increase deformation, and vary materials over a wide range of finite strain, elastodynamic, and plastic examples. Through careful benchmark ablation studies, we compare the effectiveness of HOT against seemingly plausible alternative combinations of MPM with standard multigrid and other Newton-Krylov models. We show how these alternative designs result in severe issues and poor performance. In contrast, HOT outperforms existing state-of-the-art, heavily optimized implicit MPM codes with an up to 10× performance speedup across a wide range of challenging benchmark test simulations.

We introduce a learning framework for automated floorplan generation which combines generative modeling using deep neural networks and user-in-the-loop designs to enable human users to provide sparse design constraints. Such constraints are represented by a layout graph. The core component of our learning framework is a deep neural network, Graph2Plan, which converts a layout graph, along with a building boundary, into a floorplan that fulfills both the layout and boundary constraints. Given an input building boundary, we allow a user to specify room counts and other layout constraints, which are used to retrieve a set of floorplans, with their associated layout graphs, from a database. For each retrieved layout graph, along with the input boundary, Graph2Plan first generates a corresponding raster floorplan image, and then a refined set of boxes representing the rooms. Graph2Plan is trained on RPLAN, a large-scale dataset consisting of 80K annotated floorplans. The network is mainly based on convolutional processing over both the layout graph, via a graph neural network (GNN), and the input building boundary, as well as the raster floorplan images, via conventional image convolution. We demonstrate the quality and versatility of our floorplan generation framework in terms of its ability to cater to different user inputs. We conduct both qualitative and quantitative evaluations, ablation studies, and comparisons with state-of-the-art approaches.

We present an algorithm to generate digital painting lighting effects from a single image. Our algorithm is based on a key observation: Artists use many overlapping strokes to paint lighting effects, i.e., pixels with dense stroke history tend to gather more illumination strokes. Based on this observation, we design an algorithm to both estimate the density of strokes in a digital painting using color geometry and then generate novel lighting effects by mimicking artists’ coarse-to-fine workflow. Coarse lighting effects are first generated using a wave transform and then retouched according to the stroke density of the original illustrations into usable lighting effects. Our algorithm is content-aware, with generated lighting effects naturally adapting to image structures, and can be used as an interactive tool to simplify current labor-intensive workflows for generating lighting effects for digital and matte paintings. In addition, our algorithm can also produce usable lighting effects for photographs or three-dimensional rendered images. We evaluate our approach with both an in-depth qualitative and a quantitative analysis that includes a perceptual user study. Results show that our proposed approach is not only able to produce favorable lighting effects with respect to existing approaches, but also that it is able to significantly reduce the needed interaction time.

Immersive computer graphics systems strive to generate perceptually realistic user experiences. Current-generation virtual reality (VR) displays are successful in accurately rendering many perceptually important effects, including perspective, disparity, motion parallax, and other depth cues. In this article, we introduce ocular parallax rendering, a technology that accurately renders small amounts of gaze-contingent parallax capable of improving depth perception and realism in VR. Ocular parallax describes the small amounts of depth-dependent image shifts on the retina that are created as the eye rotates. The effect occurs because the centers of rotation and projection of the eye are not the same. We study the perceptual implications of ocular parallax rendering by designing and conducting a series of user experiments. Specifically, we estimate perceptual detection and discrimination thresholds for this effect and demonstrate that it is clearly visible in most VR applications. Additionally, we show that ocular parallax rendering provides an effective ordinal depth cue and it improves the impression of realistic depth in VR.

Phased Arrays of Transducers (PATs) allow accurate control of ultrasound fields, with applications in haptics, levitation (i.e. displays) and parametric audio. However, algorithms for multi-point levitation or tactile feedback are usually limited to computing solutions in the order of hundreds of sound-fields per second, preventing the use of multiple high-speed points, a feature that can broaden the scope of applications of PATs. We present GS-PAT, a GPU multi-point phase retrieval algorithm, capable of computing 17K solutions per second for up to 32 simultaneous points in a mid-end consumer grade GPU (NVidia GTX 1660). We describe the algorithm and compare it to state of the art multi-point algorithms used for ultrasound haptics and levitation, showing similar quality of the generated sound-fields, and much higher computation rates. We then illustrate how the shift in paradigm enabled by GS-PAT (i.e. real-time control of several high-speed points) opens new applications for PAT technologies, such as in volumetric fully coloured displays, multi-point spatio-temporal tactile feedback, parametric audio and simultaneous combinations of these modalities.

We propose a GAN(Generative Adversarial Networks)-based drawing board which takes the semantic (by segmentation) and color tone (by strokes) inputs from users and automatically generates paintings. Our approach is built on a novel and lightweight feature embedding which incorporates the colorization effects into the painting generation process. Unlike the existing GAN-based image generation models which take semantics input, our drawing board has the ability to edit the local colors after generation. Our method samples the color information from users’ strokes as extra input, then feeds it into a GAN model for conditional generation. We enable the creation of pictures or paintings with semantics and color control in real-time.

Despite their cinematic appeal, turbulent flows involving fluid-solid coupling remain a computational challenge in animation. At the root of this current limitation is the numerical dispersion from which most accurate Navier-Stokes solvers suffer: proper coupling between fluid and solid often generates artificial dispersion in the form of local, parasitic trains of velocity oscillations, eventually leading to numerical instability. While successive improvements over the years have led to conservative and detail-preserving fluid integrators, the dispersive nature of these solvers is rarely discussed despite its dramatic impact on fluid-structure interaction. In this paper, we introduce a novel low-dissipation and low-dispersion fluid solver that can simulate two-way coupling in an efficient and scalable manner, even for turbulent flows. In sharp contrast with most current CG approaches, we construct our solver from a kinetic formulation of the flow derived from statistical mechanics. Unlike existing lattice Boltzmann solvers, our approach leverages high-order moment relaxations as a key to controlling both dissipation and dispersion of the resulting scheme. Moreover, we combine our new fluid solver with the immersed boundary method to easily handle fluid-solid coupling through time adaptive simulations. Our kinetic solver is highly parallelizable by nature, making it ideally suited for implementation on single- or multi-GPU computing platforms. Extensive comparisons with existing solvers on synthetic tests and real-life experiments are used to highlight the multiple advantages of our work over traditional and more recent approaches, in terms of accuracy, scalability, and efficiency.

We present a flexible and efficient approach for generating multilegged locomotion. Our model-predictive control (MPC) system efficiently generates terrain-adaptive motions, as computed using a three-level planning approach. This leverages two commonly-used simplified dynamics models, an inverted pendulum on a cart model (IPC) and a centroidal dynamics model (CDM). Taken together, these ensure efficient computation and physical fidelity of the resulting motion. The final full-body motion is generated using a novel momentum-mapped inverse kinematics solver and is responsive to external pushes by using CDM forward dynamics. For additional efficiency and robustness, we then learn a predictive model that then replaces two of the intermediate steps. We demonstrate the rich capabilities of the method by applying it to monopeds, bipeds, and quadrupeds, and showing that it can generate a very broad range of motions at interactive rates, including banked variable-terrain walking and running, hurdles, jumps, leaps, stepping stones, monkey bars, implicit quadruped gait transitions, moon gravity, push-responses, and more.

Film-quality characters typically display highly complex and expressive facial deformation. The underlying rigs used to animate the deformations of a character's face are often computationally expensive, requiring high-end hardware to deform the mesh at interactive rates. In this paper, we present a method using convolutional neural networks for approximating the mesh deformations of characters' faces. For the models we tested, our approximation runs up to 17 times faster than the original facial rig while still maintaining a high level of fidelity to the original rig. We also propose an extension to the approximation for handling high-frequency deformations such as fine skin wrinkles. While the implementation of the original animation rig depends on an extensive set of proprietary libraries making it difficult to install outside of an in-house development environment, our fast approximation relies on the widely available and easily deployed TensorFlow libraries. In addition to allowing high frame rate evaluation on modest hardware and in a wide range of computing environments, the large speed increase also enables interactive inverse kinematics on the animation rig. We demonstrate our approach and its applicability through interactive character posing and real-time facial performance capture.

We propose a new tetrahedral meshing method, fTetWild, to convert triangle soups into high-quality tetrahedral meshes. Our method builds on the TetWild algorithm, replacing the rational triangle insertion with a new incremental approach to construct and optimize the output mesh, interleaving triangle insertion and mesh optimization. Our approach makes it possible to maintain a valid floating-point tetrahedral mesh at all algorithmic stages, eliminating the need for costly constructions with rational numbers used by TetWild, while maintaining full robustness and similar output quality. This allows us to improve on TetWild in two ways. First, our algorithm is significantly faster, with running time comparable to less robust Delaunay-based tetrahedralization algorithms. Second, our algorithm is guaranteed to produce a valid tetrahedral mesh with floating-point vertex coordinates, while TetWild produces a valid mesh with rational coordinates which is not guaranteed to be valid after floating-point conversion. As a trade-off, our algorithm no longer guarantees that all input triangles are present in the output mesh, but in practice, as confirmed by our tests on the Thingi10k dataset, the algorithm always succeeds in inserting all input triangles.

Digital drawing tools are now standard in art and design workflows. These tools offer comfort, portability, and precision as well as native integration with digital-art workflows, software, and tools. At the same time, artists continue to work with long-standing, traditional drawing tools. One feature of traditional tools, well-appreciated by many artists and lacking in digital tools, is the specific and diverse range of haptic responses provided by them. Haptic feedback in traditional drawing tools provides unique, per-tool responses that help determine the precision and character of individual strokes. In this work, we address the problem of fabricating digital drawing tools that closely match the haptic feedback of their traditional counterparts. This requires the formulation and solution of a complex, co-optimization of both digital styli and the drawing surfaces they move upon. Here, a potentially direct formulation of this optimization with numerical simulation-in-the-loop is not yet viable. As in many complex design tasks, state-of-the-art methods do not currently offer predictive modeling at rates and scales that can account for the numerous, coupled, physical behaviors governing the haptics of styli and surfaces, nor for the limitations and uncertainties inherent in their fabrication processes. To address these challenges, we propose fabrication-in-the-loop optimization. Critical to making this strategy practical we construct our objective via a Gaussian Process that does not require computing derivatives with respect to design parameters. Our Gaussian Process surrogate model then provides both function estimates and confidence intervals that guide the efficient sampling of our design space. In turn, this sampling critically reduces the numbers of fabricated examples during exploration and automatically handles exploration-exploitation trade-offs. We apply our method to fabricate drawing tools that provide a wide range of haptic feedback, and demonstrate that they are often hard for users to distinguish from their traditional drawing-tool analogs.

In this paper we explore how different interaction types (POV, cloth, visual effects, and background music) can influence the user experiences in terms of interest, immersion, fun, and ease of use. To this end, we design an interactive virtual reality fashion show system with a pipeline approach and develop it using Unity3D game engine and HTC Vive HMD. A pilot study with 12 participants shows that the four interaction types can positively affect the user experiences.

We propose a novel approach for generating paths with desired exertion properties, which can be used for delivering highly realistic and immersive virtual reality applications that help users achieve exertion goals. Given a terrain as input, our optimization-based approach automatically generates feasible paths on the terrain which users can bike to perform body training in virtual reality. The approach considers exertion properties such as the total work and the perceived level of path difficulty in generating the paths. To verify our approach, we applied it to generate paths on a variety of terrains with different exertion targets and constraints. To conduct our user studies, we built an exercise bike whose force feedback was controlled by the elevation angle of the generated path over the terrain. Our user study results showed that users found exercising with our generated paths in virtual reality more enjoyable compared to traditional exercising approaches. Their energy expenditure in biking the generated paths also matched with the specified targets, validating the efficacy of our approach.

Designing a camera motion controller that has the capacity to move a virtual camera automatically in relation with contents of a 3D animation, in a cinematographic and principled way, is a complex and challenging task. Many cinematographic rules exist, yet practice shows there are significant stylistic variations in how these can be applied. In this paper, we propose an example-driven camera controller which can extract camera behaviors from an example film clip and re-apply the extracted behaviors to a 3D animation, through learning from a collection of camera motions. Our first technical contribution is the design of a low-dimensional cinematic feature space that captures the essence of a film's cinematic characteristics (camera angle and distance, screen composition and character configurations) and which is coupled with a neural network to automatically extract these cinematic characteristics from real film clips. Our second technical contribution is the design of a cascaded deep-learning architecture trained to (i) recognize a variety of camera motion behaviors from the extracted cinematic features, and (ii) predict the future motion of a virtual camera given a character 3D animation. We propose to rely on a Mixture of Experts (MoE) gating+prediction mechanism to ensure that distinct camera behaviors can be learned while ensuring generalization. We demonstrate the features of our approach through experiments that highlight (i) the quality of our cinematic feature extractor (ii) the capacity to learn a range of behaviors through the gating mechanism, and (iii) the ability to generate a variety of camera motions by applying different behaviors extracted from film clips. Such an example-driven approach offers a high level of controllability which opens new possibilities toward a deeper understanding of cinematographic style and enhanced possibilities in exploiting real film data in virtual environments.

Rendering glinty details from specular microstructure enhances the level of realism, but previous methods require heavy storage for the high-resolution height field or normal map and associated acceleration structures. In this article, we aim at dynamically generating theoretically infinite microstructure, preventing obvious tiling artifacts, while achieving constant storage cost. Unlike traditional texture synthesis, our method supports arbitrary point and range queries, and is essentially generating the microstructure implicitly. Our method fits the widely used microfacet rendering framework with multiple importance sampling (MIS), replacing the commonly used microfacet normal distribution functions (NDFs) like ground glass distribution (GGX) by a detailed local solution, with a small amount of runtime performance overhead.