[4I1-GS-7] Language media processing:

[4I1-GS-7-05] Investigation of Numerical Integration Methods for NeRF

〇Ryo Akagi¹, Takashi Matsubara¹ (1. Graduate School of Engineering Science, Osaka University)

Keywords:NeRF, Numerical Integration Methods

Neural Radiance Field (NeRF) is a technology that uses neural networks to represent a scene space based on multiple images of a subject taken from various angles, allowing for the generation of images of the subject from any viewpoint. To generate images from new viewpoints, it is necessary to evaluate the integral along the ray from the viewpoint. In practice, this is achieved by setting multiple sampling points along the ray and discretely calculating the integral value based on the information at those coordinates. Existing methods approximate the integral value under the condition that the variables are piecewise constant, which can result in lower approximation accuracy and reduced image clarity. In this pepar, we proposed the approximation for rendering by using the Runge-Kutta method to further improve of value, and investigated the effectiveness of our proposed method in terms of the accuracy of the generated images.