SIGGRAPH 2024 Papers

Existing neural rendering-based text-to-3D-portrait generation methods typically make use of human geometry prior and diffusion models to obtain guidance. However, relying solely on geometry information introduces issues such as the Janus problem, over-saturation, and over-smoothing. We present Portrait3D , a novel neural rendering-based framework with a novel joint geometry-appearance prior to achieve text-to-3D-portrait generation that overcomes the aforementioned issues. To accomplish this, we train a 3D portrait generator, 3DPortraitGAN, as a robust prior. This generator is capable of producing 360° canonical 3D portraits, serving as a starting point for the subsequent diffusion-based generation process. To mitigate the "grid-like" artifact caused by the high-frequency information in the feature-map-based 3D representation commonly used by most 3D-aware GANs, we integrate a novel pyramid tri-grid 3D representation into 3DPortraitGAN. To generate 3D portraits from text, we first project a randomly generated image aligned with the given prompt into the pre-trained 3DPortraitGAN's latent space. The resulting latent code is then used to synthesize a pyramid tri-grid. Beginning with the obtained pyramid tri-grid , we use score distillation sampling to distill the diffusion model's knowledge into the pyramid tri-grid. Following that, we utilize the diffusion model to refine the rendered images of the 3D portrait and then use these refined images as training data to further optimize the pyramid tri-grid , effectively eliminating issues with unrealistic color and unnatural artifacts. Our experimental results show that Portrait3D can produce realistic, high-quality, and canonical 3D portraits that align with the prompt.

We delve into the physics-informed neural reconstruction of smoke and obstacles through sparse-view RGB videos, tackling challenges arising from limited observation of complex dynamics. Existing physics-informed neural networks often emphasize short-term physics constraints, leaving the proper preservation of long-term conservation less explored. We introduce Neural Characteristic Trajectory Fields, a novel representation utilizing Eulerian neural fields to implicitly model Lagrangian fluid trajectories. This topology-free, auto-differentiable representation facilitates efficient flow map calculations between arbitrary frames as well as efficient velocity extraction via auto-differentiation. Consequently, it enables end-to-end supervision covering long-term conservation and short-term physics priors. Building on the representation, we propose physics-informed trajectory learning and integration into NeRF-based scene reconstruction. We enable advanced obstacle handling through self-supervised scene decomposition and seamless integrated boundary constraints. Our results showcase the ability to overcome challenges like occlusion uncertainty, density-color ambiguity, and static-dynamic entanglements. Code and sample tests are at https://github.com/19reborn/PICT_smoke.

Creating scenes for captured motions that achieve realistic human-scene interaction is crucial for 3D animation in movies or video games. As character motion is often captured in a blue-screened studio without real furniture or objects in place, there may be a discrepancy between the planned motion and the captured one. This gives rise to the need for automatic scene layout generation to relieve the burdens of selecting and positioning furniture and objects. Previous approaches cannot avoid artifacts like penetration and floating due to the lack of physical constraints. Furthermore, some heavily rely on specific data to learn the contact affordances, restricting the generalization ability to different motions. In this work, we present a physics-based approach that simultaneously optimizes a scene layout generator and simulates a moving human in a physics simulator. To attain plausible and realistic interaction motions, our method explicitly introduces physical constraints. To automatically recover and generate the scene layout, we minimize the motion tracking errors to identify the objects that can afford interaction. We use reinforcement learning to perform a dual-optimization of both the character motion imitation controller and the scene layout generator. To facilitate the optimization, we reshape the tracking rewards and devise pose prior guidance obtained from our estimated pseudo-contact labels. We evaluate our method using motions from SAMP and PROX, and demonstrate physically plausible scene layout reconstruction compared with the previous kinematics-based method.

Existing inertial motion capture techniques use the human root coordinate frame to estimate local poses and treat it as an inertial frame by default. We argue that when the root has linear acceleration or rotation, the root frame should be considered non-inertial theoretically. In this paper, we model the fictitious forces that are non-neglectable in a non-inertial frame by an auto-regressive estimator delicately designed following physics. With the fictitious forces, the force-related IMU measurement (accelerations) can be correctly compensated in the non-inertial frame and thus Newton’s laws of motion are satisfied. In this case, the relationship between the accelerations and body motions is deterministic and learnable, and we train a neural network to model it for better motion capture. Furthermore, to train the neural network with synthetic data, we develop an IMU synthesis by simulation strategy to better model the noise model of IMU hardware and allow parameter tuning to fit different hardware. This strategy not only establishes the network training with synthetic data but also enables calibration error modeling to handle bad motion capture calibration, increasing the robustness of the system. Code is available at https://xinyu-yi.github.io/PNP/.

Steered retinal projection (SRP) is an emerging display technology that combines retinal projection and pupil steering to achieve exceptional light efficiency and a consistent viewing experience. Retinal projection enables most photons from a display projector to reach the retina, and pupil steering dynamically aligns the narrow viewing window of the retinal projection with the eye. While SRP holds considerable promise, its development has been stagnant due to a lack of understanding how human vision reacts to the dynamic steering movement of the viewing window. To delve into these areas, this study introduces the first SRP system testbed specifically designed for perceptual studies on the viewing experience of pupil steering. The testbed replicates the SRP viewing experience and offers the flexibility in adjusting several parameters including steering resolution, accuracy, and latency. We conducted two perceptual studies utilizing the testbed. The first study investigates the impact of saccadic suppression, a phenomenon that reduces visual sensitivity during rapid eye movements, on the SRP viewing experience. The second study explores the trade space between eye-tracking and pupil steering performance, providing insights into the optimal balance between these factors. Additionally, we introduce a numerical model to predict the detection probability for SRP artifacts considering the temporal characteristics of global luminance and the human vision system. This model enables a more comprehensive interpretation of user study and provides preliminary hardware requirements for SRP systems. The findings from this study offer invaluable research directions that may help determine component-level development milestones for SRP development, paving the way for the practical implementation of this promising technology.

Physics-based differentiable rendering is a key ingredient for integrating forward rendering into probabilistic inference and machine learning pipelines. As a state-of-the-art formulation for differentiable rendering, differential path integrals have enabled the development of efficient Monte Carlo estimators for both interior and boundary integrals. Unfortunately, this formulation has been designed mostly for explicit geometries like polygonal meshes.

Recently, the emergence of diffusion models has opened up new opportunities for single-view reconstruction. However, all the existing methods represent the target object as a closed mesh devoid of any structural information, thus neglecting the part-based structure, which is crucial for many downstream applications, of the reconstructed shape. Moreover, the generated meshes usually suffer from large noises, unsmooth surfaces, and blurry textures, making it challenging to obtain satisfactory part segments using 3D segmentation techniques. In this paper, we present Part123, a novel framework for part-aware 3D reconstruction from a single-view image. We first use diffusion models to generate multiview-consistent images from a given image, and then leverage Segment Anything Model (SAM), which demonstrates powerful generalization ability on arbitrary objects, to generate multiview segmentation masks. To effectively incorporate 2D part-based information into 3D reconstruction and handle inconsistency, we introduce contrastive learning into a neural rendering framework to learn a part-aware feature space based on the multiview segmentation masks. A clustering-based algorithm is also developed to automatically derive 3D part segmentation results from the reconstructed models. Experiments show that our method can generate 3D models with high-quality segmented parts on various objects. Compared to existing unstructured reconstruction methods, the part-aware 3D models from our method benefit some important applications, including feature-preserving reconstruction, primitive fitting, and 3D shape editing.

Display power consumption is an emerging concern for untethered devices. This goes double for augmented and virtual extended reality (XR) displays, which target high refresh rates and high resolutions while conforming to an ergonomically light form factor. A number of image mapping techniques have been proposed to extend battery usage. However, there is currently no comprehensive quantitative understanding of how the power savings provided by these methods compare to their impact on visual quality. We set out to answer this question. To this end, we present a perceptual evaluation of algorithms (PEA) for power optimization in XR displays (PODs). Consolidating a portfolio of six power-saving display mapping approaches, we begin by performing a large-scale perceptual study to understand the impact of each method on perceived quality in the wild. This results in a unified quality score for each technique, scaled in just-objectionable-difference (JOD) units. In parallel, each technique is analyzed using hardware-accurate power models. The resulting JOD-to-Milliwatt transfer function provides a first-of-its-kind look into tradeoffs offered by display mapping techniques, and can be directly employed to make architectural decisions for power budgets on XR displays. Finally, we leverage our study data and power models to address important display power applications like the choice of display primary, power implications of eye tracking, and more 1 .

Importance sampling techniques significantly reduce variance in physically based rendering. In this article, we propose a novel online framework to learn the spatial-varying distribution of the full product of the rendering equation, with a single small neural network using stochastic ray samples. The learned distributions can be used to efficiently sample the full product of incident light. To accomplish this, we introduce a novel closed-form density model, called the Normalized Anisotropic Spherical Gaussian mixture, that can model a complex light field with a small number of parameters and that can be directly sampled. Our framework progressively renders and learns the distribution, without requiring any warm-up phases. With the compact and expressive representation of our density model, our framework can be implemented entirely on the GPU, allowing it to produce high-quality images with limited computational resources. The results show that our framework outperforms existing neural path guiding approaches and achieves comparable or even better performance than state-of-the-art online statistical path guiding techniques.

Procedural noise is a fundamental component of computer graphics pipelines, offering a flexible way to generate textures that exhibit "natural" random variation. Many different types of noise exist, each produced by a separate algorithm. In this paper, we present a single generative model which can learn to generate multiple types of noise as well as blend between them. In addition, it is capable of producing spatially-varying noise blends despite not having access to such data for training. These features are enabled by training a denoising diffusion model using a novel combination of data augmentation and network conditioning techniques. Like procedural noise generators, the model's behavior is controllable via interpretable parameters plus a source of randomness. We use our model to produce a variety of visually compelling noise textures. We also present an application of our model to improving inverse procedural material design; using our model in place of fixed-type noise nodes in a procedural material graph results in higher-fidelity material reconstructions without needing to know the type of noise in advance. Open-sourced materials can be found at https://armanmaesumi.github.io/onenoise/

Deoccluding the hidden portions of objects in a scene is a formidable task, particularly when addressing real-world scenes. In this paper, we present a new self-supervised PArallel visible-to-COmplete diffusion framework, named PACO, a foundation model for object-level scene deocclusion. Leveraging the rich prior of pre-trained models, we first design the parallel variational autoencoder, which produces a full-view feature map that simultaneously encodes multiple complete objects, and the visible-to-complete latent generator, which learns to implicitly predict the full-view feature map from partial-view feature map and text prompts extracted from the incomplete objects in the input image. To train PACO, we create a large-scale dataset with 500k samples to enable self-supervised learning, avoiding tedious annotations of the amodal masks and occluded regions. At inference, we devise a layer-wise deocclusion strategy to improve efficiency while maintaining the deocclusion quality. Extensive experiments on COCOA and various real-world scenes demonstrate the superior capability of PACO for scene deocclusion, surpassing the state of the arts by a large margin. Our method can also be extended to cross-domain scenes and novel categories that are not covered by the training set. Further, we demonstrate the deocclusion applicability of PACO in single-view 3D scene reconstruction and object recomposition. Project page: https://liuzhengzhe.github.io/Deocclude-Any-Object.github.io/.

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.

Learning from multi-view images using neural implicit signed distance functions shows impressive performance on 3D Reconstruction of opaque objects. However, existing methods struggle to reconstruct accurate geometry when applied to translucent objects due to the non-negligible bias in their rendering function. To address the inaccuracies in the existing model, we have reparameterized the density function of the neural radiance field by incorporating an estimated constant extinction coefficient. This modification forms the basis of our innovative framework, which is geared towards highfidelity surface reconstruction and the novel-view synthesis of translucent objects. Our framework contains two stages. In the reconstruction stage, we introduce a novel weight function to achieve accurate surface geometry reconstruction. Following the recovery of geometry, the second phase involves learning the distinct scattering properties of the participating media to enhance rendering. A comprehensive dataset, comprising both synthetic and real translucent objects, has been built for conducting extensive experiments. Experiments reveal that our method outperforms existing approaches in terms of reconstruction and novel-view synthesis.

Real-world fabrics, composed of threads and yarns, often display complex stress-strain relationships, making their homogenization a challenging task for fast simulation by continuum-based models. Consequently, existing homogenized yarn-level models frequently struggle with numerical stability without line search at large time steps, forcing a trade-off between model accuracy and stability. In this paper, we propose a neural-assisted homogenized constitutive model for simulating yarn-level cloth. Unlike analytic models, a neural model is advantageous in adapting to complex dynamic behaviors, and its inherent smoothness naturally mitigates stability issues. We also introduce a sector-based warm-start strategy to accelerate the data collection process in homogenization. This model is trained using collected strain energy datasets and its accuracy is validated through both qualitative and quantitative experiments. Thanks to our model’s stability, our simulator can now achieve two-orders-of-magnitude speedups with large time steps compared to previous models.

We introduce a novel neural network-based computational pipeline as a representation-agnostic slicer for multi-axis 3D printing. This advanced slicer can work on models with diverse representations and intricate topology. The approach involves employing neural networks to establish a deformation mapping, defining a scalar field in the space surrounding an input model. Isosurfaces are subsequently extracted from this field to generate curved layers for 3D printing. Creating a differentiable pipeline enables us to optimize the mapping through loss functions directly defined on the field gradients as the local printing directions. New loss functions have been introduced to meet the manufacturing objectives of support-free and strength reinforcement. Our new computation pipeline relies less on the initial values of the field and can generate slicing results with significantly improved performance.

The idea of using a neural network to represent continuous vector fields (i.e., neural fields) has become popular for solving PDEs arising from physics simulations. Here, the classical spatial discretization (e.g., finite difference) of PDE solvers is replaced with a neural network that models a differentiable function, so the spatial gradients of the PDEs can be readily computed via autodifferentiation. When used in fluid simulation, however, neural fields fail to capture many important phenomena, such as the vortex shedding experienced in the von Kármán vortex street experiment. We present a novel neural network representation for fluid simulation that augments neural fields with explicitly enforced boundary conditions as well as a Monte Carlo pressure solver to get rid of all weakly enforced boundary conditions. Our method, the Neural Monte Carlo method (NMC), is completely mesh-free, i.e., it doesn’t depend on any grid-based discretization. While NMC does not achieve the state-of-the-art accuracy of the well-established grid-based methods, it significantly outperforms previous mesh-free neural fluid methods on fluid flows involving intricate boundaries and turbulence regimes.

Recent work on using neural fields to represent surfaces has resulted in significant improvements in representational capability and computational efficiency. However, to our knowledge, most existing work has focused on implicit representations such as signed distance fields or volumes, and little work has explored their application to discrete surface geometry, i.e., 3D meshes, limiting the applicability of neural surface representations.

Gaussian scale spaces are a cornerstone of signal representation and processing, with applications in filtering, multiscale analysis, anti-aliasing, and many more. However, obtaining such a scale space is costly and cumbersome, in particular for continuous representations such as neural fields. We present an efficient and lightweight method to learn the fully continuous, anisotropic Gaussian scale space of an arbitrary signal. Based on Fourier feature modulation and Lipschitz bounding, our approach is trained self-supervised, i.e., training does not require any manual filtering. Our neural Gaussian scale-space fields faithfully capture multiscale representations across a broad range of modalities, and support a diverse set of applications. These include images, geometry, light-stage data, texture anti-aliasing, and multiscale optimization.

This paper presents a method to leverage arbitrary neural network architecture for control variates. Control variates are crucial in reducing the variance of Monte Carlo integration, but they hinge on finding a function that both correlates with the integrand and has a known analytical integral. Traditional approaches rely on heuristics to choose this function, which might not be expressive enough to correlate well with the integrand. Recent research alleviates this issue by modeling the integrands with a learnable parametric model, such as a neural network. However, the challenge remains in creating an expressive parametric model with a known analytical integral. This paper proposes a novel approach to construct learnable parametric control variates functions from arbitrary neural network architectures. Instead of using a network to approximate the integrand directly, we employ the network to approximate the anti-derivative of the integrand. This allows us to use automatic differentiation to create a function whose integration can be constructed by the antiderivative network. We apply our method to solve partial differential equations using the Walk-on-sphere algorithm. Our results indicate that this approach is unbiased and uses various network architectures to achieve lower variance than other control variate methods.