Why the coarse stage of sampling were sent to the network for inference twice

[YZsZY]

Hello kwea, sorry for disturbing you! I would like to ask a question.

I noticed that the fine stage merged the sample points from the coarse stage and sent them to the network for inference, here I have a doubt, because the sample points from the coarse stage have already been inferred before, why do they have to be sent to the network for inference again here, is it to facilitate the final volume rendering? But it feels like this will add a lot of computation

nerf_pl/models/rendering.py

Line 229 in 52aeb38

z_vals, _ = torch.sort(torch.cat([z_vals, z_vals_], -1), -1)