Guim

You need to install threads package. `luarocks install threads`

I have found the origin of the problem. https://github.com/soumith/dcgan.torch/blob/e0578020ecd5c1b5df6356e7059b2b46670ae5a8/main.lua#L240 Generator is saved as `netG:clearState()`. If saved without `clearState()`, images on evaluation mode are generated properly.

It doesn't make much of a difference. You would only simply give the discriminator one training iteration extra training compared to the generator. And in fact, you are interested on...

Hi again, I misunderstood your question, sorry! When you say "then call backward on netD in fGx", you are refering [to this line of code](https://github.com/soumith/dcgan.torch/blob/master/main.lua#L200), right? `local df_dg = netD:updateGradInput(input,...

This time I think I got your point. I agree with you. Then you would simply change [this line](https://github.com/soumith/dcgan.torch/blob/master/main.lua#L197): `local output = netD.output -- netD:forward(input) was already executed in fDx,...