KiloNeRF: Speeding up Neural Radiance Fields with Thousands of Tiny MLPs
NeRF synthesizes novel views of a scene with unprecedented quality by fitting a neural radiance field to RGB images. However, NeRF requires querying a deep Multi-Layer Perceptron (MLP) millions of times, leading to slow rendering times, even on modern GPUs. In this paper, we demonstrate that significant speed-ups are possible by utilizing thousands of tiny MLPs instead of one single large MLP. In our setting, each individual MLP only needs to represent parts of the scene, thus smaller and faster-to-evaluate MLPs can be used. By combining this divide-and-conquer strategy with further optimizations, rendering is accelerated by two orders of magnitude compared to the original NeRF model without incurring high storage costs. Further, using teacher-student distillation for training, we show that this speed-up can be achieved without sacrificing visual quality.
https://creiser.github.io/kilonerf/
Видео KiloNeRF: Speeding up Neural Radiance Fields with Thousands of Tiny MLPs канала Andreas Geiger
https://creiser.github.io/kilonerf/
Видео KiloNeRF: Speeding up Neural Radiance Fields with Thousands of Tiny MLPs канала Andreas Geiger
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
Learning Robust Driving PoliciesDALI 2018 - Ingmar Posner: Driving Autonomy - Learning to Perceive and to ActKING: Generating Safety-Critical Driving Scenarios for Robust Imitation via Kinematics GradientsTracking by DetectionCounterfactual Latent DanceGenerating Images and 3D ShapesARAH: Animatable Volume Rendering of Articulated Human SDFsSLIM: Self-Supervised LiDAR Scene Flow and Motion SegmentationDALI 2018 - Olivier Bousquet: What could a theory of Deep Learning look like?RVC 2020 - Keynote: Ross Girshick - Robustness Across the Data Abundance SpectrumExploiting Object Similarity in 3D ReconstructionLIBELAS: Efficient Large-Scale Stereo MatchingLocally Aware Piecewise Transformation Fields for 3D Human Mesh RegistrationConstraining 3D Fields for Reconstruction and View SynthesisKITTI-360: A Novel Dataset and Benchmarks for Urban Scene Understanding in 2D and 3DExploring Data Aggregation in Policy Learning for Vision-based Urban Autonomous DrivingDisplets: Resolving Stereo Ambiguities using Object KnowledgeManmohan Chandraker: Physically-Motivated Learning of Shape, Material and Lighting in Complex ScenesJoint 3D Object and Layout Inference from a single RGB-D Image3D Deep Learning In Function SpaceLabel Efficient Visual Abstractions for Autonomous Driving