Real/Fake Accuracy

[Queuecumber]

If I want to compute the real/fake accuracy of the discriminator for a batch using this method, do I need to do:

real_accuracy = sigmoid(real_logits - mean(fake_logits)) >= 0.5
fake_accuracy = sigmoid(fake_logits - mean(real_logits)) < 0.5

e.g. the same input that I would give to the BCE with logits loss goes to the sigmoid

Or do I do the standard GAN thing and compute

real_accuracy = sigmoid(real_logits) >= 0.5
fake_accuracy = sigmoid(fake_logits) < 0.5

[AlexiaJM]

There's no concept of accuracy in the traditional sense here, but if you want something as close possible with RaSGAN, your first approach is how I would do it. The second approach won't make sense.

[Queuecumber]

Thanks, usually when I train a GAN I look for the accuracy to sit around ~80% for both real and fake to know if it's doing ok, is there a surrogate metric you'd recommend for the RA loss?