Efficiency of storing lists of weights and ray_samples

[yhaddouda]

I am trying to profile the nerfstudio implementation of nerfacto, and I have a question. I have noticed that in the propsal sampler they store weights and raysamples in a list that contains for every step three elements. And these two lists are stored in the outputs dictionnary at the end. Although they say the saving only happens in training mode because they require memory and take time, I do not understand why do we need to save those two lists and calculate them other than debugging purposes, so is there another reason that I am missing for saving those two heavy lists ?