_loss_g_masked_cond divided by batch_size

[kgu3]

Hi Albert,
Very nice job!!

I got a question about the loss for AU.
When computing the loss for Discriminator's AU regression head, you divided it by batch_size, as shown below:
self._loss_g_masked_cond = self._criterion_D_cond(d_fake_desired_img_masked_cond, self._desired_cond) / self._B * self._opt.lambda_D_cond

Just wondering is there any reason for that?

Thank you!