Understanding 3D Gaussian Splatting

The object being represented

A 3D Gaussian can be described by a center, covariance, opacity and appearance attributes. The covariance determines the local spatial extent and orientation of the primitive.

Why covariance matters

Rather than treating every point as infinitesimal, the Gaussian gives each primitive an anisotropic support region. This makes the representation both continuous and differentiable.

What I want to explore next

I am interested in whether Gaussian representations can be adapted to scientific 3D fields where observations are dense in some dimensions and extremely sparse in others.