Browsing by Author "Kurz, Andreas"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    DONeRF: Towards Real-Time Rendering of Compact Neural Radiance Fields using Depth Oracle Networks
    (The Eurographics Association and John Wiley & Sons Ltd., 2021) Neff, Thomas; Stadlbauer, Pascal; Parger, Mathias; Kurz, Andreas; Mueller, Joerg H.; Chaitanya, Chakravarty R. Alla; Kaplanyan, Anton S.; Steinberger, Markus; Bousseau, Adrien and McGuire, Morgan
    The recent research explosion around implicit neural representations, such as NeRF, shows that there is immense potential for implicitly storing high-quality scene and lighting information in compact neural networks. However, one major limitation preventing the use of NeRF in real-time rendering applications is the prohibitive computational cost of excessive network evaluations along each view ray, requiring dozens of petaFLOPS. In this work, we bring compact neural representations closer to practical rendering of synthetic content in real-time applications, such as games and virtual reality. We show that the number of samples required for each view ray can be significantly reduced when samples are placed around surfaces in the scene without compromising image quality. To this end, we propose a depth oracle network that predicts ray sample locations for each view ray with a single network evaluation. We show that using a classification network around logarithmically discretized and spherically warped depth values is essential to encode surface locations rather than directly estimating depth. The combination of these techniques leads to DONeRF, our compact dual network design with a depth oracle network as its first step and a locally sampled shading network for ray accumulation. With DONeRF, we reduce the inference costs by up to 48x compared to NeRF when conditioning on available ground truth depth information. Compared to concurrent acceleration methods for raymarching-based neural representations, DONeRF does not require additional memory for explicit caching or acceleration structures, and can render interactively (20 frames per second) on a single GPU.
  • No Thumbnail Available
    Item
    Efficient Rendering of Participating Media for Multiple Viewpoints
    (The Eurographics Association, 2023) Stojanovic, Robert; Weinrauch, Alexander; Tatzgern, Wolfgang; Kurz, Andreas; Steinberger, Markus; Bikker, Jacco; Gribble, Christiaan
    Achieving realism in modern games requires the integration of participating media effects, such as fog, dust, and smoke. However, due to the complex nature of scattering and partial occlusions within these media, real-time rendering of high-quality participating media remains a computational challenge. To address this challenge, traditional approaches of real-time participating media rendering involve storing temporary results in a view-aligned grid before ray marching through these cached values. In this paper, we investigate alternative hybrid worldand view-aligned caching methods that allow for the sharing of intermediate computations across cameras in a scene. This approach is particularly relevant for multi-camera setups, such as stereo rendering for VR and AR, local split-screen games, or cloud-based rendering for game streaming, where a large number of players may be in the same location. Our approach relies on a view-aligned grid for near-field computations, which enables us to capture high-frequency shadows in front of a viewer. Additionally, we use a world-space caching structure to selectively activate distant computations based on each viewer's visibility, allowing for the sharing of computations and maintaining high visual quality. The results of our evaluation demonstrate computational savings of up to 50% or more, without compromising visual quality.