Show simple item record

dc.contributor.authorNeff, Thomasen_US
dc.contributor.authorStadlbauer, Pascalen_US
dc.contributor.authorParger, Mathiasen_US
dc.contributor.authorKurz, Andreasen_US
dc.contributor.authorMueller, Joerg H.en_US
dc.contributor.authorChaitanya, Chakravarty R. Allaen_US
dc.contributor.authorKaplanyan, Anton S.en_US
dc.contributor.authorSteinberger, Markusen_US
dc.contributor.editorBousseau, Adrien and McGuire, Morganen_US
dc.description.abstractThe 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.en_US
dc.publisherThe Eurographics Association and John Wiley & Sons Ltd.en_US
dc.subjectComputing methodologies --> Rendering
dc.titleDONeRF: Towards Real-Time Rendering of Compact Neural Radiance Fields using Depth Oracle Networksen_US
dc.description.seriesinformationComputer Graphics Forum
dc.description.sectionheadersNeural Rendering

Files in this item


This item appears in the following Collection(s)

  • 40-Issue 4
    Rendering 2021 - Symposium Proceedings

Show simple item record