search

nerfstudio-project / gsplat

CUDA accelerated rasterization of gaussian splatting
https://docs.gsplat.studio/
Apache License 2.0
2.26k stars 289 forks source link

Isotropic Gaussians? #438

Open ImNotPrepared opened 1 month ago

ImNotPrepared commented 1 month ago

Hi,

I want to use isotropic gaussians rendering, is there any fast way to achieve that? Appreciate your kind help.

Best

maturk commented 1 month ago

Hi. I think it should be possible to make scales = torch.nn.Parameter(torch.rand(N,1)) for size (N,1) (or make them size dist_avg which is knn neighbour avg, as in original paper, check simple_trainer.py) and when calling rasterization(...) function call with rasterization(..., scales = scales.repeat(1,3), ...) to reshape them to (N,3) but identical values hence isotropic. For each gaussian you optimize single scale only.