Question about the chosen proxy mesh of each 2D Gaussian

[Yang-Xijie]

You mention that you choose icosahedron as the proxy mesh of each 2D Gaussian.

..., we follow 3DGRT to fit a bounding proxy of each 2D Gaussian using an adaptive icosahedron mesh...

However, it seems simplier to use only two triangles to bound the 2D Gaussian, as in EnvGS (https://zju3dv.github.io/envgs/):

..., we propose to represent each 2D Gaussian with two triangles...

This will only use 2 triangles per primitive instead of 20 triangles used by icosahedron. I wonder if you regard this a better choice over icosahedron?

[SuLvXiangXin]

I follow 3DGRT to use a icosahedron to bound a Gaussian, this might be a tighter bound than using 2 triangles to represent a 2D Gaussian, which means more Gaussians will be intersected by the rays in EnvGS. However, their results should be the same, and it is more likely EnvGS is faster. I will try it when I'm available.