Waymo urban dataset

[antithing]

Hi, might this implementation work on the waymo urban dataset?

thanks!

[guanjunwu]

I'm not sure.
Reconstruction urban scene such as Waymo dataset always need miliions of 3D Gaussians, computing the deformation of all the 3D Gaussians by Gaussian Deformation Network is expensive, but using motion mask to split dynamic and static part may work.
And I think Street Gaussians and pvg may have better performance.

[huster-wugj]

I'm not sure. Reconstruction urban scene such as Waymo dataset always need miliions of 3D Gaussians, computing the deformation of all the 3D Gaussians by Gaussian Deformation Network is expensive, but using motion mask to split dynamic and static part may work. And I think Street Gaussians and pvg may have better performance.

I encountered the same problem. In urban scenes, especially the ground with less texture, the effect is very poor. Do you have any good suggestions? Will 4dgs be less effective at reconstructing large scenes than scenes with boundaries?My colleague has tried street gaussian, but the effect is not ideal in weakly textured scenes.