local variable 'z' referenced before assignment about dcgan HOT 4 OPEN

z is just noise.
You can simply add:
z = np.random.normal(loc=0.0, scale=1.0, size=[flags.batch_size, flags.z_dim]).astype(np.float32)
before
Gen.save_weights('{}/Gen.npz'.format(flags.checkpoint_dir), format='npz')

from dcgan.

I'm having the same issue. Did you end up fixing it?

from dcgan.

how is it possible? the z is generated during training

from dcgan.

I think this is caused by z first being assigned in the block that iterates over input images. If your input directory has no images, rather than reporting this as an error, it'll go right to reference z and fail on that.

from dcgan.