A problem about your code

[q145492675]

https://github.com/King-Of-Knights/Keras-ACGAN-CIFAR10/blob/c5a89becb4e5a56dbf34bf7c45516bbf6c451d36/cifar10.py#L253

Hello sir.

I have a problem with these part of your code. Based on you previous codes, you set Real data has label in the range of 0.7-1.2 and Fake data is 0.0-0.3. But why in this place you reverse them and let Real data is 0.0-1.3?

[King-Of-Knights]

It is still 0.0-0.3 not 0.0-1.3 and It has been explained above.

make the labels the noisy for the discriminator: occasionally flip the labels

refer to https://github.com/soumith/ganhacks/blob/master/README.md trick 6 for more detail. Actually, I believe it is somehow a trick to combat with local optima, but as you can see in this code, label will be reversed after 30 times normal trainning. This trick won't make a different to the result.

[q145492675]

Oh, I have read this Doc before. But this trick of help you jump out of the local optima is interesting, haha. It makes sense : )

[King-Of-Knights]

I strongly trust the quality could be improved, but I really have no idea how to do this. any suggestion will be appreciated!