Mesh + residual GS, How to optimize?

[kaikai23]

Hi, this work is so cool!

May I ask a question regarding how the "mesh+residualGS" works during training?
(1) Do you have a special renderer that can produce a rendered image from the hybrid representation, or do you render them separately and somehow blend them together?
(2) Is the mesh part somehow differentiable, so that we can even optimize them? I guess it is not trivial, but would like to get some insights whether it is possible or what would be the difficulties, thank you!