search

kcheng1021 / GaussianPro

[ICML2024] Official code for GaussianPro: 3D Gaussian Splatting with Progressive Propagation
https://kcheng1021.github.io/gaussianpro.github.io/
MIT License
551 stars 36 forks source link

Can lidar point clouds be used for further deep supervision in the waymo dataset? #35

Open Yellowshuohahaha opened 2 months ago

kcheng1021 commented 2 months ago

Hi, I am not sure the meaning of 'deep supervision'. Lidar could provide extra dense points. However, for in the wild videos, the lidar seems inaccessible.

Yellowshuohahaha commented 2 months ago

@kcheng1021 Thank you for your reply. What I mean is, will using lidar generated depth maps for depth supervision be better than using SFM and MVS generated depth maps? Have you done any experiments on this? Because lidar typically has better geometric accuracy, although it has sparsity.

kcheng1021 commented 2 months ago

Thanks for your advice. I haven’t tried to compare lidar point clouds yet, because we focus on in the wild scenes, not just autonomous driving. It will be so nice for you to try it and share the results. I believe that the Lidar point cloud will also improve 3DGS. I will also try to add this comparison later.

Yellowshuohahaha commented 2 months ago

Thank you！I will try it and share the experimental results！