Where is the pre-training mentioned in your paper?

[eveneveno]

Are you suggesting the pre-training stage which is the grid-branch only training mentioned in the paper? Practically we activate the nerf branch training with the conig parameter add_nerf