search

graphdeco-inria / gaussian-splatting

Original reference implementation of "3D Gaussian Splatting for Real-Time Radiance Field Rendering"
https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/
Other
13.63k stars 1.76k forks source link

Cloned Points Not Moving Along Gradient Direction #904

Open Metro1998 opened 1 month ago

Metro1998 commented 1 month ago

Hi, First of all, thank you very much for open-sourcing such excellent work. I've been working on a point cloud densification process, and I noticed that during the cloning of points, the new points are not being moved along the gradient direction. But you have mentioned in the paper

For this, it is preferable to clone the Gaussians, by simply creating a copy of the same size, and moving it in the direction of the positional gradient.

Here is the relevant part of the code: https://github.com/graphdeco-inria/gaussian-splatting/blob/472689c0dc70417448fb451bf529ae532d32c095/scene/gaussian_model.py#L374C5-L388C1

As you can see, the cloned points (new_xyz) are just the same as the original points. They are not moved in the direction of the gradient. Is this final implementation what you originally intended?

torracxiaokeai commented 1 month ago

I have same question with you... and i find the answer in other issue https://github.com/graphdeco-inria/gaussian-splatting/issues/217#issuecomment-1727102619