Is anyone know what enforces the generated images to be at the correct range of [0,1]?
You could try to use a sigmoid
at the end of your generator.
In the DCGAN example a tanh
is used to get predictions in [-1, 1]
, which might work better.
Is anyone know what enforces the generated images to be at the correct range of [0,1]?
You could try to use a sigmoid
at the end of your generator.
In the DCGAN example a tanh
is used to get predictions in [-1, 1]
, which might work better.