Abstract Virtual Reality (VR) applications aim to create an immersive virtual world, which demands a high level of visual realism. The analytical material models commonly used in VR often fall short of reproducing complex real‐world appearances. Recently, neural materials have emerged as a promising alternative, offering a compact yet effective representation of real‐world materials. Deploying neural materials on low‐power mobile VR devices poses significant challenges due to the computational complexity of neural networks and the high display resolution and frame rate requirements of VR devices (commonly 72+ frames per second). We address these challenges by leveraging texture‐space shading with spatiotemporal computation amortization, driven by a compact, coarse‐to‐fine neural material model of extremely low capacity. Thanks to our distillation training scheme, our compact neural materials achieve visual quality comparable to NeuMIP [KMX*21] at a much lower cost. Our method reaches over 90 FPS on a mobile VR device (Meta Quest 3) even under multiple light sources.

Measured Bidirectional Texture Function (BTF) can faithfully reproduce a realistic appearance but is costly to acquire and store due to its 6D nature (2D spatial and 4D angular). Therefore, it is practical and necessary for rendering to synthesize BTFs from a small example patch. While previous methods managed to produce plausible results, we find that they seldomly take into consideration the property of being dynamic, so a BTF must be synthesized before the rendering process, resulting in limited size, costly pre-generation and storage issues. In this paper, we propose a dynamic BTF synthesis scheme, where a BTF at any position only needs to be synthesized when being queried. Our insight is that, with the recent advances in neural dimension reduction methods, a BTF can be decomposed into disjoint low-dimensional components. We can perform dynamic synthesis only on the positional dimensions, and during rendering, recover the BTF by querying and combining these low-dimensional functions with the help of a lightweight Multilayer Perceptron (MLP). Consequently, we obtain a fully dynamic 6D BTF synthesis scheme that does not require any pre-generation, which enables efficient rendering of our infinitely large and non-repetitive BTFs on the fly. We demonstrate the effectiveness of our method through various types of BTFs taken from UBO2014 [Weinmann et al. 2014].

Abstract We present a new software ray tracing solution that efficiently computes visibilities in dynamic scenes. We first introduce a novel scene representation: ray‐aligned occupancy map array (ROMA) that is generated by rasterizing the dynamic scene once per frame. Our key contribution is a fast and low‐divergence tracing method computing visibilities in constant time, without constructing and traversing the traditional intersection acceleration data structures such as BVH. To further improve accuracy and alleviate aliasing, we use a spatiotemporal scheme to stochastically distribute the candidate ray samples. We demonstrate the practicality of our method by integrating it into a modern real‐time renderer and showing better performance compared to existing techniques based on distance fields (DFs). Our method is free of the typical artifacts caused by incomplete scene information, and is about 2.5×–10× faster than generating and tracing DFs at the same resolution and equal storage.

Basis functions provide both the abilities for compact representation and the properties for efficient computation. Therefore, they are pervasively used in rendering to perform all-frequency shading. However, common basis functions, including spherical harmonics (SH), wavelets, and spherical Gaussians (SG) all have their own limitations, such as low-frequency for SH, not rotationally invariant for wavelets, and no multiple product support for SG. In this paper, we present neural basis functions, an implicit and data-driven set of basis functions that circumvents the limitations with all desired properties. We first introduce a representation neural network that takes any general 2D spherical function (e.g. environment lighting, BRDF, and visibility) as input and projects it onto the latent space as coefficients of our neural basis functions. Then, we design several lightweight neural networks that perform different types of computation, giving our basis functions different computational properties such as double/triple product integrals and rotations. We demonstrate the practicality of our neural basis functions by integrating them into all-frequency shading applications, showing that our method not only achieves a compression rate of and 10 × -40 × better performance than wavelets at equal quality, but also renders all-frequency lighting effects in real-time without the aforementioned limitations from classic basis functions.

Complex luminaires, such as grand chandeliers, can be extremely costly to render because the light-emitting sources are typically encased in complex refractive geometry, creating difficult light paths that require many samples to evaluate with Monte Carlo approaches. Previous work has attempted to speed up this process, but the methods are either inaccurate, require the storage of very large lightfields, and/or do not fit well into modern path-tracing frameworks. Inspired by the success of deep networks, which can model complex relationships robustly and be evaluated efficiently, we propose to use a machine learning framework to compress a complex luminaire's lightfield into an implicit neural representation. Our approach can easily plug into conventional renderers, as it works with the standard techniques of path tracing and multiple importance sampling (MIS). Our solution is to train three networks to perform the essential operations for evaluating the complex luminaire at a specific point and view direction, importance sampling a point on the luminaire given a shading location, and blending to determine the transparency of luminaire queries to properly composite them with other scene elements. We perform favorably relative to state-of-the-art approaches and render final images that are close to the high-sample-count reference with only a fraction of the computation and storage costs, with no need to store the original luminaire geometry and materials.

Abstract Gradient‐domain rendering can highly improve the convergence of light transport simulation using the smoothness in image space. These methods generate image gradients and solve an image reconstruction problem with rendered image and the gradient images. Recently, a previous work proposed a gradient‐domain volumetric photon density estimation for homogeneous participating media. However, the image reconstruction relies on traditional L1 reconstruction, which leads to obvious artifacts when only a few rendering passes are performed. Deep learning based reconstruction methods have been exploited for surface rendering, but they are not suitable for volume density estimation. In this paper, we propose an unsupervised neural network for image reconstruction of gradient‐domain volumetric photon density estimation, more specifically for volumetric photon mapping, using a variant of GradNet with an encoded shift connection and a separated auxiliary feature branch, which includes volume based auxiliary features such as transmittance and photon density. Our network smooths the images on global scale and preserves the high frequency details on a small scale. We demonstrate that our network produces a higher quality result, compared to previous work. Although we only considered volumetric photon mapping, it's straightforward to extend our method for other forms, like beam radiance estimation.