convert H3DGS model (.hier) to 3DGS legacy model (.ply)

[Yang-Xijie]

As described in README,

The interface includes a field for tau (size limit) which defines the desired granularity setting. Note that tau = 0 will try to render the complete dataset (all leaf nodes).

Is it reasonable to export all the finest gaussians from the final result (.hier) and save them into a legacy 3DGS model (.ply)? Will there be any differences if I compare the rendering results of them (using H3DGS viewer and 3DGS viewer respectively)?

How to filter all leaf nodes by setting tau = 0 in Python code?

[Yang-Xijie]

related issues:

• Can we get a merged .py file? #18
• How can I see a good result with .PLY file? #29
• Can you explain the output file merged.hier? #79

[Yang-Xijie]

For a single chunk, gaussians in the point_cloud.ply are the leaves in the .hier file.
However, for a merged large scene with several chunks, since there is no such merged.ply, leaves can only be exported from the merged.hier file.

[ameuleman]

Hi,
Getting the indices of the Gaussians to render with a zero threshold should give you the leaves (see expand_to_size).
We can extract the relevant Gaussians from the indices: scene.gaussians._xyz = scene.gaussians._xyz[render_inds] (this has to run for all other 3D Gaussian parameters).
Then we should be able to save the leaves with save_ply.
Note that we have not tested this process and it will require a lot of GPU memory to load the full scene without hierarchy.

[ameuleman]

Also, the viewer has to support correct antialiasing (see #47). This is now available in 3DGS.