Clarification regarding scale initialization

[deivse]

Hi, I have a question regarding the scale initialization - there are the following lines at the end of EDGSTrainer.init_with_corr:

with torch.no_grad():
            gaussians =  self.gaussians
            gaussians._scaling =  gaussians.scaling_inverse_activation(gaussians.scaling_activation(gaussians._scaling)*0.5)

Is my understanding correct that this is equivalent to setting a 2x lower scaling_factor in configs/train.yaml? Is there any reason this is done as the last step?