In research on making the rendering of learned radiance fields (e.g., neural radiance fields) faster, many approaches use Spherical Harmonics (SH) to approximate view-dependent color (e.g. baking SH into classical 3D structures such as meshes with textures, or point clouds).

Historically, it seems that SH have predominantly been used to represent diffuse lighting. Why are they now also employed for view-dependent color?

  • $\begingroup$ I think this question is good and it can be even more specific. SH is good at approximating low frequency lighting since low degree SH leads to simple analytical approximation (diffusion approximation) and low memory consumption. I am also interested in why plenoxels and 3D GS only use 2 degree SH to achieve photo-realistic effects even for specular components, as it is always said that high frequency effect can not be handled well with low degree SH. Is is because of alpha blending or interpolation? $\endgroup$ Nov 2, 2023 at 1:34


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.