remove mapping network in training stage

[dragen1860]

Dear all
I wonder what if i remove the mapping network and only take style vector s from ref. any one give some tips or did some experiments? thank you.

[yl4579]

@980202006 has done it, you may comment on that issue. It is as simple as removing the lines that train with the mapping network and removing the style diversification loss to prevent loss divergence.