SIGGRAPH 2020 Papers

The human visual system uses numerous cues for depth perception, including disparity, accommodation, motion parallax and occlusion. It is incumbent upon virtual-reality displays to satisfy these cues to provide an immersive user experience. Multifocal displays, one of the classic approaches to satisfy the accommodation cue, place virtual content at multiple focal planes, each at a different depth. However, the content on focal planes close to the eye do not occlude those farther away; this deteriorates the occlusion cue as well as reduces contrast at depth discontinuities due to leakage of the defocus blur. This paper enables occlusion-aware multifocal displays using a novel ConeTilt operator that provides an additional degree of freedom --- tilting the light cone emitted at each pixel of the display panel. We show that, for scenes with relatively simple occlusion configurations, tilting the light cones provides the same effect as physical occlusion. We demonstrate that ConeTilt can be easily implemented by a phase-only spatial light modulator. Using a lab prototype, we show results that demonstrate the presence of occlusion cues and the increased contrast of the display at depth edges.

We introduce the first neural optimization framework to solve a classical instance of the tiling problem. Namely, we seek a non-periodic tiling of an arbitrary 2D shape using one or more types of tiles---the tiles maximally fill the shape's interior without overlaps or holes. To start, we reformulate tiling as a graph problem by modeling candidate tile locations in the target shape as graph nodes and connectivity between tile locations as edges. Further, we build a graph convolutional neural network, coined TilinGNN, to progressively propagate and aggregate features over graph edges and predict tile placements. TilinGNN is trained by maximizing the tiling coverage on target shapes, while avoiding overlaps and holes between the tiles. Importantly, our network is self-supervised, as we articulate these criteria as loss terms defined on the network outputs, without the need of ground-truth tiling solutions. After training, the runtime of TilinGNN is roughly linear to the number of candidate tile locations, significantly outperforming traditional combinatorial search. We conducted various experiments on a variety of shapes to showcase the speed and versatility of TilinGNN. We also present comparisons to alternative methods and manual solutions, robustness analysis, and ablation studies to demonstrate the quality of our approach.

Establishing a robust measure for material similarity that correlates well with human perception is a long-standing problem. A recent work presented a deep learning model trained to produce a feature space that aligns with human perception by gathering human subjective measures. The resulting metric outperforms objective existing ones. In this work, we aim to understand whether this increased performance is a result of using human perceptual data or is due to the nature of feature learnt by deep learning models. We train similar networks with objective measures (BRDF similarity or classification task) and show that these networks can predict human judgements as well, suggesting that the non-linear features learnt by convolutional network might be a key to model material perception.

Previous research in pattern formation using reaction-diffusion mostly focused on static domains, either for computational simplicity or mathematical tractability. In this work, we have explored the expressiveness of combining simple mechanisms as a possible explanation for pigmentation pattern formation, where tissue growth plays a crucial role. Our motivation is not only to realistically reproduce natural patterns but also to get insights into the underlying biological processes. Therefore, we present a novel approach to generate realistic animal skin patterns. First, we describe the approximation of tissue growth by a series of discrete matrix expansion operations. Then, we combine it with an adaptation of Turing's non-linear reaction-diffusion model, which enforces upper and lower bounds to the concentrations of the involved chemical reagents. We also propose the addition of a single-reagent continuous autocatalytic reaction, called reinforcement, to provide a mechanism to maintain an already established pattern during growth. By careful adjustment of the parameters and the sequencing of operations, we closely match the appearance of a few real species. In particular, we reproduce in detail the distinctive features of the leopard skin, also providing a hypothesis for the simultaneous productions of the most common melanin types, eumelanin and pheomelanin.

Interacting with people across large distances is important for remote work, interpersonal relationships, and entertainment. While such face-to-face interactions can be achieved using 2D video conferencing or, more recently, virtual reality (VR), telepresence systems currently distort the communication of eye contact and social gaze signals. Although methods have been proposed to redirect gaze in 2D teleconferencing situations to enable eye contact, 2D video conferencing lacks the 3D immersion of real life. To address these problems, we develop a system for face-to-face interaction in VR that focuses on reproducing photorealistic gaze and eye contact. To do this, we create a 3D virtual avatar model that can be animated by cameras mounted on a VR headset to accurately track and reproduce human gaze in VR. Our primary contributions in this work are a jointly-learnable 3D face and eyeball model that better represents gaze direction and upper facial expressions, a method for disentangling the gaze of the left and right eyes from each other and the rest of the face allowing the model to represent entirely unseen combinations of gaze and expression, and a gaze-aware model for precise animation from headset-mounted cameras. Our quantitative experiments show that our method results in higher reconstruction quality, and qualitative results show our method gives a greatly improved sense of presence for VR avatars.

We describe the design and evolution of UberBake, a global illumination system developed by Activision, which supports limited lighting changes in response to certain player interactions. Instead of relying on a fully dynamic solution, we use a traditional static light baking pipeline and extend it with a small set of features that allow us to dynamically update the precomputed lighting at run-time with minimal performance and memory overhead. This means that our system works on the complete set of target hardware, ranging from high-end PCs to previous generation gaming consoles, allowing the use of lighting changes for gameplay purposes. In particular, we show how to efficiently precompute lighting changes due to individual lights being enabled and disabled and doors opening and closing. Finally, we provide a detailed performance evaluation of our system using a set of production levels and discuss how to extend its dynamic capabilities in the future.

We present a method to render virtual touch, such that the stimulus produced by a tactile device on a user's skin matches the stimulus computed in a virtual environment simulation. To achieve this, we solve the inverse mapping from skin stimulus to device configuration thanks to a novel optimization algorithm. Within this algorithm, we use a device-skin simulation model to estimate rendered stimuli, we account for trajectory-dependent effects efficiently by decoupling the computation of the friction state from the optimization of device configuration, and we accelerate computations using a neural-network approximation of the device-skin model. Altogether, we enable real-time tactile rendering of rich interactions including smooth rolling, but also contact with edges, or frictional stick-slip motion. We validate our algorithm both qualitatively through user experiments, and quantitatively on a BioTac biomimetic finger sensor.

Members of the blind and visually impaired community rely heavily on tactile illustrations - raised line graphics on paper that are felt by hand - to understand geometric ideas in school textbooks, depict a story in children's books, or conceptualize exhibits in museums. However, these illustrations often fail to achieve their goals, in large part due to the lack of understanding in how 3D shapes can be represented in 2D projections. This paper describes a new technique to design tactile illustrations considering the needs of blind individuals. Successful illustration design of 3D objects presupposes identification and combination of important information in topology and geometry. We propose a twofold approach to improve shape understanding. First, we introduce a part-based multi-projection rendering strategy to display geometric information of 3D shapes, making use of canonical viewpoints and removing reliance on traditional perspective projections. Second, curvature information is extracted from cross sections and embedded as textures in our illustrations.

We present a novel linear subdivision scheme for face-based tangent directional fields on triangle meshes. Our subdivision scheme is based on a novel coordinate-free representation of directional fields as halfedge-based scalar quantities, bridging the mixed finite-element representation with discrete exterior calculus. By commuting with differential operators, our subdivision is structure preserving: it reproduces curl-free fields precisely and reproduces divergence-free fields in the weak sense. Moreover, our subdivision scheme directly extends to directional fields with several vectors per face by working on the branched covering space. Finally, we demonstrate how our scheme can be applied to directional-field design, advection, and robust earth mover’s distance computation, for efficient and robust computation.

Scattering from specular surfaces produces complex optical effects that are frequently encountered in realistic scenes: intricate caustics due to focused reflection, multiple refraction, and high-frequency glints from specular microstructure. Yet, despite their importance and considerable research to this end, sampling of light paths that cause these effects remains a formidable challenge. In this article, we propose a surprisingly simple and general sampling strategy for specular light paths including the above examples, unifying the previously disjoint areas of caustic and glint rendering into a single framework. Given two path vertices, our algorithm stochastically finds a specular subpath connecting the endpoints. In contrast to prior work, our method supports high-frequency normal- or displacement-mapped geometry, samples specular-diffuse-specular ("SDS") paths, and is compatible with standard Monte Carlo methods including unidirectional path tracing. Both unbiased and biased variants of our approach can be constructed, the latter often significantly reducing variance, which may be appealing in applied settings (e.g. visual effects). We demonstrate our method on a range of challenging scenes and evaluate it against state-of-the-art methods for rendering caustics and glints.

Efficiently rendering direct lighting from millions of dynamic light sources using Monte Carlo integration remains a challenging problem, even for off-line rendering systems. We introduce a new algorithm---ReSTIR---that renders such lighting interactively, at high quality, and without needing to maintain complex data structures. We repeatedly resample a set of candidate light samples and apply further spatial and temporal resampling to leverage information from relevant nearby samples. We derive an unbiased Monte Carlo estimator for this approach, and show that it achieves equal-error 6×-60× faster than state-of-the-art methods. A biased estimator reduces noise further and is 35×-65× faster, at the cost of some energy loss. We implemented our approach on the GPU, rendering complex scenes containing up to 3.4 million dynamic, emissive triangles in under 50 ms per frame while tracing at most 8 rays per pixel.

In this paper, we introduce a numerical technique to generate sample distributions in arbitrary dimension for improved accuracy of Monte Carlo integration. We point out that optimal transport offers theoretical bounds on Monte Carlo integration error, and that the recently-introduced numerical framework of sliced optimal transport (SOT) allows us to formulate a novel and efficient approach to generating well-distributed high-dimensional pointsets. The resulting sliced optimal transport sampling, solely involving repeated 1D solves, is particularly simple and efficient for the common case of a uniform density over a d-dimensional ball. We also construct a volume-preserving map from a d-ball to a d-cube (generalizing the Shirley-Chiu mapping to arbitrary dimensions) to offer fast SOT sampling over d-cubes. We provide ample numerical evidence of the improvement in Monte Carlo integration accuracy that SOT sampling brings compared to existing QMC techniques, and derive a projective variant for rendering which rivals, and at times outperforms, current sampling strategies using low-discrepancy sequences or optimized samples.

We introduce a novel deep learning framework for data-driven motion retargeting between skeletons, which may have different structure, yet corresponding to homeomorphic graphs. Importantly, our approach learns how to retarget without requiring any explicit pairing between the motions in the training set. We leverage the fact that different homeomorphic skeletons may be reduced to a common primal skeleton by a sequence of edge merging operations, which we refer to as skeletal pooling. Thus, our main technical contribution is the introduction of novel differentiable convolution, pooling, and unpooling operators. These operators are skeleton-aware, meaning that they explicitly account for the skeleton's hierarchical structure and joint adjacency, and together they serve to transform the original motion into a collection of deep temporal features associated with the joints of the primal skeleton. In other words, our operators form the building blocks of a new deep motion processing framework that embeds the motion into a common latent space, shared by a collection of homeomorphic skeletons. Thus, retargeting can be achieved simply by encoding to, and decoding from this latent space. Our experiments show the effectiveness of our framework for motion retargeting, as well as motion processing in general, compared to existing approaches. Our approach is also quantitatively evaluated on a synthetic dataset that contains pairs of motions applied to different skeletons. To the best of our knowledge, our method is the first to perform retargeting between skeletons with differently sampled kinematic chains, without any paired examples.

We propose a new light-weight face capture system capable of reconstructing both high-quality geometry and detailed appearance maps from a single exposure. Unlike currently employed appearance acquisition systems, the proposed technology does not require active illumination and hence can readily be integrated with passive photogrammetry solutions. These solutions are in widespread use for 3D scanning humans as they can be assembled from off-the-shelf hardware components, but lack the capability of estimating appearance. This paper proposes a solution to overcome this limitation, by adding appearance capture to photogrammetry systems. The only additional hardware requirement to these solutions is that a subset of the cameras are cross-polarized with respect to the illumination, and the remaining cameras are parallel-polarized. The proposed algorithm leverages the images with the two different polarization states to reconstruct the geometry and to recover appearance properties. We do so by means of an inverse rendering framework, which solves per texel diffuse albedo, specular intensity, and high-resolution normals, as well as global specular roughness considering the subsurface scattering nature of skin. We show results for a variety of human subjects of different ages and skin typology, illustrating how the captured fine-detail skin surface and subsurface scattering effects lead to realistic renderings of their digital doubles, also in different illumination conditions.

Digital cameras can only capture a limited range of real-world scenes' luminance, producing images with saturated pixels. Existing single image high dynamic range (HDR) reconstruction methods attempt to expand the range of luminance, but are not able to hallucinate plausible textures, producing results with artifacts in the saturated areas. In this paper, we present a novel learning-based approach to reconstruct an HDR image by recovering the saturated pixels of an input LDR image in a visually pleasing way. Previous deep learning-based methods apply the same convolutional filters on wellexposed and saturated pixels, creating ambiguity during training and leading to checkerboard and halo artifacts. To overcome this problem, we propose a feature masking mechanism that reduces the contribution of the features from the saturated areas. Moreover, we adapt the VGG-based perceptual loss function to our application to be able to synthesize visually pleasing textures. Since the number of HDR images for training is limited, we propose to train our system in two stages. Specifically, we first train our system on a large number of images for image inpainting task and then fine-tune it on HDR reconstruction. Since most of the HDR examples contain smooth regions that are simple to reconstruct, we propose a sampling strategy to select challenging training patches during the HDR fine-tuning stage. We demonstrate through experimental results that our approach can reconstruct visually pleasing HDR results, better than the current state of the art on a wide range of scenes.

In the field of computational design of meta materials, complex patterns of meso-structures have seen increased interest because of the ability to locally control their flexibility through adjustment of the meso-structure parameters. Such structures come with a number of advantages like, despite the simplicity of their fabrication, their ability to nestle to sophisticated free-form surfaces. However, the simulation of such complex structures still comes with a high computational cost. We propose an approach to reduce this computational cost by abstracting the meso-structures and encoding the properties of their elastic deformation behavior into a different set of material parameters. We can thus obtain an approximation of the deformed pattern by simulating a simplified version of the pattern using the computed material parameters.

Frictional contacts are the primary way by which physical bodies interact, yet they pose many numerical challenges. Previous works have devised robust methods for handling collisions in elastic bodies, cloth, or fiber assemblies such as hair, but the performance of many of those algorithms degrades when applied to objects with different topologies or constitutive models, or simply cannot scale to high-enough numbers of contacting points. In this work we propose a unified approach, able to handle a large class of dynamical objects, that can solve for millions of contacts with unbiased Coulomb friction while keeping computation time and memory usage reasonable. Our method allows seamless coupling between the various simulated components that comprise virtual characters and their environment. Furthermore, our proposed approach is simple to implement and can be easily integrated in popular time integrators such as Projected Newton or ADMM.

Visual design tasks often involve tuning many design parameters. For example, color grading of a photograph involves many parameters, some of which non-expert users might be unfamiliar with. We propose a novel user-in-the-loop optimization method that allows users to efficiently find an appropriate parameter set by exploring such a high-dimensional design space through much easier two-dimensional search subtasks. This method, called sequential plane search, is based on Bayesian optimization to keep necessary queries to users as few as possible. To help users respond to plane-search queries, we also propose using a gallery-based interface that provides options in the two-dimensional subspace arranged in an adaptive grid view. We call this interactive framework Sequential Gallery since users sequentially select the best option from the options provided by the interface. Our experiment with synthetic functions shows that our sequential plane search can find satisfactory solutions in fewer iterations than baselines. We also conducted a preliminary user study, results of which suggest that novices can effectively complete search tasks with Sequential Gallery in a photo-enhancement scenario.