What is it about?
With a sparse, mobile, and easy-to-deploy capture setup, LitNeRF performs relightable rendering by decomposing the scene radiance into interpretable components that are motivated by physically-based rendering. In simple terms, LitNeRF makes the best-performing generative AI based rendering technique, i.e. NeRF, relightable.
Featured Image
Photo by Mason Kimbarovsky on Unsplash
Why is it important?
While the AI based methods, NeRFs, are state-of-the-art for rendering, they cannot be relit in different lighting environment (sun, sky, indoor, etc.) in a straight-forward way. LitNeRF provides an elegant extension of NeRF that can be used for rendering with different illumination. It also provides interpretable components such as diffuse, specular, direct and indirect light transport factors that are important in computer graphics. Moreover, LitNeRF requires a capture setup that is an order of magnitude smaller than the existing methods for relighting.
Perspectives
In this work, all of the intrinsic radiance fields (diffuse, specular, direct, indirect light transport, etc.) are simple MLPs similar to that of NeRF, and they are combined together using the standard rendering equation. It was very interesting to see that unsupervised intrinsic decomposition of the radiance of NeRF is possible using the ideas of physically-based rendering.
Kripasindhu Sarkar
Read the Original
This page is a summary of: LitNeRF: Intrinsic Radiance Decomposition for High-Quality View Synthesis and Relighting of Faces, December 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3610548.3618210.
You can read the full text:
Contributors
The following have contributed to this page
