Critic Loss Function

Course Q&A Generative Adversarial Networks (GANS) Build Basic Generative Adversarial Networks
1

The loss function is given by : min(g)max(c) crit_real_pred - crit_fake_pred + gp
my question : why did we do crit_fake_pred - crit_real_pred+ gp for critic loss

2

@theLifter welcome to the community. You have correctly stated the loss function as

min(g)max(c) crit_real_pred - crit_fake_pred + gp

however while asking the question you have reserved the two as

crit_fake_pred - crit_real_pred

which is incorrect. As a very high level intuition, we want to count real predictions positively and fake predictions negatively. Hence the equation. Hope it helps.

3

crit_fake_pred - crit_real_pred : This was done in the assignment while calculating critic loss function. So, I was just confirming.

Thanks.

5

This is answered here by Wendy Why is the Generator Loss in WGAN negative mean of the predicted image - #5 by Wendy

1 Like