Open ToTheBeginning opened 6 years ago
Setting the imageSize from 64 to 32 breaks the convolutional layers, because the kernel sizes don't fit anymore!
you can change the kernel size = 2 for final layer of the discriminator and kernel size = 1 for generator. This will work for imagesize = 32
After run about 15 epoch, the generator seems to collapse to a parameter setting where it always emits the same point.
Did anyone run the script successfully?