How do you compute loss in later two phase?
Johnson-yue opened this issue · comments
Hi, I read your paper , but I don't understand how to compute loss in later two phase? I know the G1_1 , G1_2, G1_3 is the same network but different phase.
In first phase:
X_fake1 = G1_1(E, M, Xgt)
and computing loss.....
but in three phase:
X_fake3 = G1_3(E, X_fake2)
and computing loss...... How I got X_fake2?
I guess
method 1:
X_fake_2 = G1_2(E, Cgt)
X_fake3 = G1_3(E, X_fake2)
and computing loss......
method 2:
torch.no_grad():
X_fake_2 = G1_2(E, Cgt)
X_fake3 = G1_3(E, X_fake2)
and computing loss......
@Johnson-yue The three phases are not end-to-end, u can only start the next phase when the phase before is converged. So u can just save the model of well trained generator G1_2 to treat it as a fixed generator, and X_fake2 is generated from it and feed it to train a new G1_2 into G1_3.
ok, I see