def build_cnn_for_recon(batch_size=10):
# output cnn recon images(with alias) from the undersample k-space im
parser = argparse.ArgumentParser()
parser.add_argument('--lr',
metavar='float',
nargs=1,
default=['0.001'],
help='initial learning rate')
parser.add_argument('--l2',
metavar='float',
nargs=1,
default=['1e-6'],
help='l2 regularisation')
args = parser.parse_args()
Nx, Ny = 128, 128
# Specify network
input_shape = (batch_size, 2, Nx, Ny)
# Comnstruct and Load D5-C5 with pretrained params
net_config, net, = build_d5_c5(input_shape)
with np.load('./models/pretrained/d5_c5.npz') as f:
param_values = [f['arr_{0}'.format(i)] for i in range(len(f.files))]
lasagne.layers.set_all_param_values(net, param_values)
# Compile function
train_fn, val_fn = compile_fn(net, net_config, args)
return val_fn
def load_network(input_shape, network_path=network_path):
"""
Loads the pretrained network, specified by the absolute_path.
:param input_shape: - (batch_size, 2, Nx, Ny)
:param network_path: - path to network weights
:returns: net_config, network
"""
#input_shape = (batch_size, 2, Nx, Ny)
#net_config, net, = build_d2_c2(input_shape)
# Load D5-C5 with pretrained params
net_config, network, = build_d5_c5(input_shape)
# D5-C5 with pre-trained parameters
with np.load(network_path) as f:
param_values = [f['arr_{0}'.format(i)] for i in range(len(f.files))]
lasagne.layers.set_all_param_values(network, param_values)
return net_config, network
save_every = 5
# Configure directory info
project_root = '.'
save_dir = join(project_root, 'models/%s' % model_name)
if not os.path.isdir(save_dir):
os.makedirs(save_dir)
# Create dataset
n_train, n_validate, n_test = 160, 40, 56
train, validate, test = create_dummy_data([n_train, n_validate, n_test])
# Specify network
input_shape = (batch_size, 2, Nx, Ny)
# net_config, net, = build_d2_c2(input_shape)
net_config, net, = build_d5_c5(input_shape)
# D5-C5 with pre-trained parameters
# with np.load('./models/pretrained/d5_c5.npz') as f:
# param_values = [f['arr_{0}'.format(i)] for i in range(len(f.files))]
# lasagne.layers.set_all_param_values(net, param_values)
print('Undersampling Rate: {:.2f}'.format(1 / acc))
# Compile function
train_fn, val_fn = compile_fn(net, net_config, args)
print('Start Training...')
for epoch in xrange(num_epoch):
t_start = time.time()