About SDS loss takes normal as input

[VLadImirluren]

thanks for your good reproduction!!
I noticed that when exp_iter_ratio <= self.opt.latent_iter_ratio, the shading ='normal'
Then, color = (normal + 1) 2,
WHY IT WORKS ?! It's amazing.
why it convergence to a better place, why it won't be meanless(because the NeRF is randomly init)
thanks again!