for filter in visibility_filters:
opacities_new[filter] = self._opacity[filter]
As opacities_new is initialized by inverse_sigmoid(0.4) , but self.get_opacity is sigmoid(self._opacity) in range of (0,1).
Did I misunderstand anything?
By the way, the function reset_opacity_nonvisible runs each 2000 mapping iterations. However, in 3DGS, the opacity only resets in the initial iteration. I am wondering why MonoGS needs to reset the opacity of nonvisible Gaussians. It seems not explained in the paper. Thanks in advance!
It seems to have a problem with the function
reset_opacity_nonvisible
on this line https://github.com/muskie82/MonoGS/blob/6c9254c319d8bff5caeef65259e6bb0941a9b9f6/gaussian_splatting/scene/gaussian_model.py#L373 I think this line should be:As
opacities_new
is initialized byinverse_sigmoid(0.4)
, butself.get_opacity
issigmoid(self._opacity)
in range of (0,1). Did I misunderstand anything?By the way, the function
reset_opacity_nonvisible
runs each 2000 mapping iterations. However, in 3DGS, the opacity only resets in the initial iteration. I am wondering why MonoGS needs to reset the opacity of nonvisible Gaussians. It seems not explained in the paper. Thanks in advance!