Question about batch size

[gkouros]

Hi.

I see that in the main training loop you use a varying batch size, rather than a fixed one. Can you please explain the rationale behind this decision and how the update rule works?

Thanks.

[half-potato]

It's the same thing NerfAcc does. The idea is to maximize the batch size while limiting the number of ray samples so you don't run out of memory.

[gkouros]

Ok. Thanks for the explanation.