Some problems with the initial point cloud

[moqiyinlun]

Since the pre_n3d.py script does not handle points3D.txt in the code, how can I get the initial point cloud for each frame to train?

[lizhan17]

you can get sparse points with the function here

from thirdparty.gaussian_splatting.helper3dg import getcolmapsinglen3d

[moqiyinlun]

Thank you very much for your answer, and then I have a small question if the code handle masked images (white background) as in the original Gaussian with the args "--white_background"? Thanks again!

[lizhan17]

In this repo along with the three dataset, we opt not to use white pixels as background.

[moqiyinlun]

Thanks! I get pretty good results!