How to get the Gaussian visualization in the demo?

[yifliu3]

Hi! The demo results are truly remarkable, and I was thoroughly impressed by the quality of the reconstructions.

I'm particularly interested in the 3D Gaussian primitives used in your method. Is there a way to visualize these reconstructed Gaussians?

[LiHaodong0217]

you can use any viewer to visualize 3dgs render result which support to render 3dgs.
maybe you can try it:https://jatentaki.github.io/assets/gaussianviewer/index.html

[CanCanZeng]

Hi @LiHaodong0217 @yifliu3 , I also interested in this project and I'd like to ask how to save the gaussians in gaussian_splatting format (ply point cloud file)?

[LiHaodong0217]

so you need to find a function about "save_ply" like "save_ply(scene,'gf.ply')"
then feed the "xx.ply" to application like SuperSplat(https://playcanvas.com/supersplat/editor)

[haofeixu]

Hi, you can try to use this code https://github.com/donydchen/mvsplat/blob/main/src/paper/generate_point_cloud_figure_mvsplat.py to visualize the 3D Gaussians.
It would take a bit of time for me to clean my code due to the project deadlines, thank you for your patience.

[CanCanZeng]

I found an export_ply function, but it requires inputs of scales and rotations, and depthsplat only outputs covariance in the Gaussians class. I couldn't find a function that converts covariance to scales and rotations. Why predict covariance directly instead of scales and rotations?