Details about noise buffer

[Orig1n]

I'm confused when I read the code in w_plus_projector.py. I notice the noise buffer is optimized during the inversion, but you didn't save it to the npy or the pkl. How can the generator generate correct videos with random noises?