def __init__(self, data_shape):
self.data_shape = data_shape
self.discriminator = None
self.generator = None
self.adversarial = None
self.define_gan()
self.noisy_samples = NoiseMaker(generator=self.generator)
self.performance_output_path = 'performance/temp/'
if not os.path.exists(self.performance_output_path):
os.makedirs(self.performance_output_path)
def __init__(self, data_shape):
"""
Initialize SiDN-GAN
"""
self.data_shape = data_shape
self.discriminator = None
self.generator = None
self.adversarial = None
self.define_gan()
self.noise_maker = NoiseMaker(shape=self.data_shape, noise_type='s&p')
self.performance_output_path = 'performance/siamese_dn_gan_' + str(
datetime.now().date())
# Warm up
x = 0
for i in range(BASE_ITERATIONS):
x += 1
op = "dirichlet.rvs() (scipy frozen distribution, size 9)"
import scipy.stats # noqa
d = scipy.stats.dirichlet([.2] * 9)
with time_operation(op, BASE_ITERATIONS) as op:
for i in range(op.num_interations):
d.rvs()
from noise_maker import NoiseMaker # noqa
NOISE_MAKER = NoiseMaker(1000)
op = "NOISE_MAKER.make_noise(.2, 10)"
with time_operation(op, BASE_ITERATIONS) as op:
for i in range(op.num_interations):
NOISE_MAKER.make_noise(.2, 10)
op = "random.randint(0, 9999)"
with time_operation(op, BASE_ITERATIONS) as op:
for i in range(op.num_interations):
random.randint(0, 9999)
op = "[0.0 for x in agents]"
agents = [0, 1]
with time_operation(op, BASE_ITERATIONS) as op:
for i in range(op.num_interations):
s = [0.0 for x in agents]
generated = generator.predict(noisy)
# save the generator model
model_file = path + '/model_%04d.h5' % (epoch + 1)
generator.save(model_file)
fig_file = path + '/plot_%04d' % ((epoch + 1))
measure_and_plot(original_images=test_data,
noisy_images=noisy,
generated_images=generated,
path=fig_file)
print('>Saved model and figures to', path)
if __name__ == '__main__':
dataset = Dataset(dataset='caltech256')
dataset.split_test_data(test_sample=2000)
noise_maker = NoiseMaker(shape=dataset.data_shape, noise_type='s&p')
model_folder = 'C:/PycharmProjects/NeuralNetworks-GAN/performance/caltech256-128x128-siamese_dn_gan_2019-12-21'
for epoch in range(20):
generator_path = model_folder + '/epoch-%04d' % (
epoch + 1) + '/model_%04d.h5' % (epoch + 1)
generator = load_model(generator_path)
performance(generator, noise_maker, epoch, dataset.test_data,
model_folder)
def __post_init__(self):
super().__post_init__()
self.noise_maker = NoiseMaker(1000)
if self.policy_overrides is None:
self.policy_overrides = [None, None]