Bayesian NeRF Training

Modeling scene-centered images as mixture of Gaussian distributions, we limit the number of training observations required for neural radiance field training by selecting to minimize uncertainty.