About the number of parameters

[qiqiing]

For image super-resolution task，are all the parameters to be optimized just N latent vectors and N adaptive channel importance weights?Do parameters such as convolution weights in the pre-trained PGGAN model used not need to be updated with specific tasks?