Fourier Feature Networks and Neural Volume Rendering

Matthew Johnson

Microsoft Research, Cambridge, UK

Abstract

Fourier Feature Networks are an exciting new development in Computer Vision, and their use for modeling radiance fields has produced a range of impressive results at the meeting point of Computer Vision and Computer Graphics. In this session Matthew will cover the motivation behind using Fourier features in neural network training, introduce the fundamentals of volumetric ray casting, and then show how we can use Fourier Feature Networks to render high-quality novel views of complex 3D scenes. The session will also incorporate a workshop in using Azure Machine Learning to train large neural rendering models of the kind described in the lecture.