- Team: TaiDenG
- Project: NeRF in ray tracing
This project is in the early stages of development.
Neural radiance field (NeRF) is a state-of-the-art novel view synthesis method capable of rendering photorealistic scenes by only training with collective images and camera pose.
However, NeRF uses an implicit modeling style that inputs a position with viewing direction and outputs the color and density of that position, which can not be used as a general object file that contains geometric vertices.
As a result, the traditional pipeline of a path tracer is not able to do ray tracing for a learned NeRF model. But, there is a workaround that you can extract the geometric and color from NeRF using methods like marching cubes so that it can be compatible with a standard ray tracing renderer. Unfortunately, the extraction is not perfect, and the extracted model will have lower quality than the original NeRF.
This project aims to integrate the NeRF model into a ray tracer so that the NeRF can be directly rendered and participate in ray tracing without previously extracting the geometric and color.