General Guide on Hyperparameters

[Chumsy0725]

Hello @Shiwei-Ren,

Thank you very much for the pretty cool work.

I notice there's a set of hyperparameters used in FastGS, and in train.sh, there are some slight changes to these parameters in the MipNeRF360 scene. I was wondering if you could provide some general guideline how to use these parameters for custom scenes to get the best results out of them. For instance, Indoor Vs Outdoor/Bounded Vs Unbounded Scenes?

Thank you!

[Shiwei-Ren]

Hello! Thank you for your interest, and sorry for the late reply. Here is an explanation regarding the adjustment of the hyperparameters with clear patterns: loss_thresh, grad_abs_thresh, and grad_thresh affect the number of Gaussians—generally, the lower the values, the more Gaussians are generated, which improves rendering quality but increases training time. Reducing mult can relatively decrease training time.