the label of generated DCGAN pictures

[qiaoguan]

hi~, In your paper, for the generated DCGAN pictures, the probability for each class is the same ,i.e. 1/k（k is the numer of the class）, while in the training stage , we have to assign a label for any training sample, so i do not know how to assign label for the generated DCGAN pictures, however, i looked at your source code in prepare_gan_data.m, it seems that the label for the generated DCGAN samples is 0, i'm a little confused, can you explaining why? thanks! （chinese is also OK if it is easier to explain）

[layumi]

Hi @qiaoguan,
Sorry for the late response.
Yes, the input label for the generated images is 0 (when preparing dataset).
In fact, I write the "1/k" in the loss function.
If the input label is 0, the code know the input is generated sample and then assign a different loss.