Browsing by Author "Stadlbauer, Pascal"

Now showing 1 - 3 of 3
  • No Thumbnail Available
    Item
    Clouds in the Cloud: Efficient Cloud-Based Rendering of Real-Time Volumetric Clouds
    (The Eurographics Association, 2023) Weinrauch, Alexander; Lorbek, Stephan; Tatzgern, Wolfgang; Stadlbauer, Pascal; Steinberger, Markus; Bikker, Jacco; Gribble, Christiaan
    Volumetric clouds play a crucial role in creating realistic, dynamic, and immersive virtual outdoor environments. However, rendering volumetric clouds in real-time presents a significant computational challenge on end-user devices. In this paper, we investigate the viability of moving computations to remote servers in the cloud and sharing them among many viewers in the same virtual world, without compromising the perceived quality of the final renderings. We propose an efficient rendering method for volumetric clouds and cloud shadows utilizing caches placed in the cloud layers and directly on the surface of objects. Volumetric cloud properties, like density and lightning, are cached on spheres positioned to represent cloud layers at varying heights. Volumetric cloud shadows are cached directly on the surfaces of receiving objects. This allows efficient rendering in scenarios where multiple viewers observe the same cloud formations by sharing redundant calculations and storing them over multiple frames. Due to the placement and structure of our caches, viewers on the ground still perceive plausible parallax under movement on the ground. In a user study, we found that viewers hardly perceive quality reductions even when computations are shared for viewers that are hundreds of meters apart. Due to the smoothness of the appearance of clouds, caching structures can use significantly reduced resolution and as such allow for efficient rendering even in single-viewer scenarios. Our quantitative experiments demonstrate computational cost savings proportional to the number of viewers placed in the scene when relying on our caches compared to traditional rendering.
  • 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
    Surface Light Cones: Sharing Direct Illumination for Efficient Multi-viewer Rendering
    (The Eurographics Association, 2023) Stadlbauer, Pascal; Weinrauch, Alexander; Tatzgern, Wolfgang; Steinberger, Markus; Bikker, Jacco; Gribble, Christiaan
    Even though stochastic methods and hardware supported ray tracing are increasingly used for computing direct illumination, the efficient real-time rendering of dynamic area light sources still forms a challenge. In this paper, we propose a method for representing and caching direct illumination information using a compact multi-cone representation that is stored on the surface of objects. While shading due to direct illumination is typically heavily view-dependent, the incoming radiance for surface points is view-independent. Relying on cones, to represent the projection of the dominant visible light sources, allows to reuse the incoming radiance information across frames and even among multiple cameras or viewers within the same scene. Progressively refining and updating the cone structures not only allows to adapt to dynamic scenes, but also leads to reduced noise levels in the output images compared to sampling based methods. Relying on surface light cones allows to render single viewer setups 2-3x faster than random sampling, and 1.5-2x faster than reservoir-based sampling with the same quality. The main selling point for surface light cones is multi-camera rendering, For stereo rendering, our approach essentially halves the time required for determining direct light visibility. For rendering in the cloud, where multiple viewers are positioned close to another, such as in virtual meetings, gathering locations in games, or online events such as virtual concerts, our approach can reduce overall rendering times by a factor of 20x for as few as 16 viewers in a scene compared to traditional light sampling. Finally, under heavily constraint ray budgets where noise levels typically overshadow bias, surface light cones can dramatically reduce noise.