Anttwo/SuGaR

about joint optimization

Opened this issue · 0 comments

Hello, thanks for your awesome work.
I've got a question about the joint optimization of mesh and gaussians.

Intuitively, vertices of the mesh will serve as a good initialization of the position of gaussians.
But in table 4,5 vanilla 3DGS which does not use mesh shows better rendering performance than SUGAR.

What do you think the reason for this is?

Thank you again for your wonderful work.