search

baldFemale / beautyGAN-tf-Implement

100 stars 21 forks source link

Wrong discreminator #4

Open GuWenyuan opened 5 years ago

GuWenyuan commented 5 years ago

Discreminators used on fake_A and fake_B are wrong. (Line 233, 234 in main.py) fake_A is supposed to be after-makeup face and fake_B is non-makeup. So according to the paper, we should use d_A for fake_B and d_B for fake_A. Simply switch parameters "d_A" and "d_B" on line 233 and 234 should work.

lynlindasy commented 4 years ago

But I think fake_A in his code means fake and no-makeup img generated from real makeup img B, and fake_B means fake makeup img generated from real no-makeup img A. And I think the code at line 225 in main.py demonstrates my opinion. He used fake_B to get the first result of the generator and used fake_A to get the second result. Am I wrong? Waiting for you reply.

Discreminators used on fake_A and fake_B are wrong. (Line 233, 234 in main.py) fake_A is supposed to be after-makeup face and fake_B is non-makeup. So according to the paper, we should use d_A for fake_B and d_B for fake_A. Simply switch parameters "d_A" and "d_B" on line 233 and 234 should work.