DCGAN tutorial - generator does not converge

Hi, I am trying to generate trees from an images dataset I prepared. As a tool, I am building on the DCGAN tutorial. During the training, the loss function of the generator does not converge (see image below). Do you have any suggestion on how to fix this? I already tried changing the learning rate from 0.0002 to 001, but the behavior of the loss function did not change. Thanks for the help!


Are you resetting some parameters after one epoch or are you now shuffling the real dataset?
The loss spikes look a bit weird to me.