Code questions.. or Typos?

[denniswittich]

These two questions came up, when I went threw the code:

• In line 127 of pokeGan.py the bn1 tensor is created. But actually it is never used.. instead the next line continues with lrelu(conv1,...). Any reason for that?

• Why is tanh used to generate the final images (line 113). Shouldn't the images values be between 0 and 1? So would not the sigmoid activation be the better choice here?

Any answers / ideas?

[todd-koeck]

I have the same question about bn1, but I think its just left over from a cut-n-paste as the generator has the same code.

I could be completely wrong on this though as I didn't get good results on the included pokemon images nor am I getting much better results on some other data I've thrown at it. I believe GAN networks take a lot of time and data to train, so there might just be a need to train more, or there could be problems in the code.

On tanh vs. sigmoid, I understand (in my very limited knowledge) that tanh trains better in many cases than sigmoid.