def test_reconstruction_fista_fft2(self):
""" Test all the registered transformations.
"""
print("Process test FFT2 FISTA::")
for image in self.images:
fourier = FFT2(samples=convert_mask_to_locations(
fftshift(self.mask)),
shape=image.shape)
data = fourier.op(image.data)
fourier_op = FFT2(convert_mask_to_locations(fftshift(self.mask)),
shape=image.shape)
print("Process test with image '{0}'...".format(
image.metadata["path"]))
for nb_scale in self.nb_scales:
print("- Number of scales: {0}".format(nb_scale))
for name in self.names:
print(" Transform: {0}".format(name))
linear_op = WaveletN(wavelet_name=name, nb_scale=4)
gradient_op = GradSynthesis2(data=data,
fourier_op=fourier_op,
linear_op=linear_op)
prox_op = Threshold(0)
x_final, transform, _, _ = sparse_rec_fista(
gradient_op=gradient_op,
linear_op=linear_op,
prox_op=prox_op,
cost_op=None,
lambda_init=1.0,
max_nb_of_iter=self.nb_iter,
verbose=0)
fourier_0 = FFT2(samples=convert_mask_to_locations(
fftshift(self.mask)),
shape=image.shape)
data_0 = fourier_0.op(numpy.fft.fftshift(image.data))
self.assertTrue(
numpy.allclose(x_final.any(),
numpy.fft.ifftshift(
fourier_0.adj_op(data_0)).any(),
rtol=1e-10))
mean_square_error = numpy.mean(
numpy.abs(
x_final -
numpy.fft.ifftshift(fourier_0.adj_op(data_0)))**2)
print(" Mean Square Error = ", mean_square_error)
def reco_wav(kspace,
gradient_op,
mu=1 * 1e-8,
max_iter=10,
nb_scales=4,
wavelet_name='db4'):
# for now this is only working with my fork of pysap-fastMRI
# I will get it changed soon so that we don't need to ask for a specific
# pysap-mri install
from ..wavelets import WaveletDecimated
from mri.numerics.reconstruct import sparse_rec_fista
linear_op = WaveletDecimated(
nb_scale=nb_scales,
wavelet_name=wavelet_name,
padding='periodization',
)
prox_op = LinearCompositionProx(
linear_op=linear_op,
prox_op=SparseThreshold(Identity(), None, thresh_type="soft"),
)
gradient_op.obs_data = kspace
cost_op = None
x_final, _, _, _ = sparse_rec_fista(
gradient_op=gradient_op,
linear_op=Identity(),
prox_op=prox_op,
cost_op=cost_op,
xi_restart=0.96,
s_greedy=1.1,
mu=mu,
restart_strategy='greedy',
pov='analysis',
max_nb_of_iter=max_iter,
metrics=None,
metric_call_period=1,
verbose=0,
progress=False,
)
x_final = np.abs(x_final)
x_final = crop_center(x_final, 320)
return x_final
gradient_op, linear_op, prox_op, cost_op = generate_operators(
data=kspace_obs,
wavelet_name="BsplineWaveletTransformATrousAlgorithm",
samples=kspace_loc,
nb_scales=4,
non_cartesian=True,
uniform_data_shape=image.shape,
gradient_space="synthesis")
# Start the FISTA reconstruction
max_iter = 20
x_final, transform, costs, metrics = sparse_rec_fista(gradient_op,
linear_op,
prox_op,
None,
mu=1e-9,
lambda_init=1.0,
max_nb_of_iter=max_iter,
atol=1e-4,
verbose=1)
image_rec = pysap.Image(data=np.abs(x_final))
image_rec.show()
#############################################################################
# Condata-Vu optimization
# -----------------------
#
# We now want to refine the zero order solution using a Condata-Vu
# optimization.
# Here no cost function is set, and the optimization will reach the
# maximum number of iterations. Fill free to play with this parameter.
# ------------------
#
# We now want to refine the zero order solution using a FISTA optimization.
# The cost function is set to Proximity Cost + Gradient Cost
# Generate operators
gradient_op, linear_op, prox_op, cost_op = generate_operators(
data=kspace_obs,
wavelet_name="sym8",
samples=kspace_loc,
mu=6 * 1e-7,
nb_scales=4,
non_cartesian=True,
uniform_data_shape=image.shape,
gradient_space="synthesis")
# Start the FISTA reconstruction
max_iter = 200
x_final, costs, metrics = sparse_rec_fista(gradient_op=gradient_op,
linear_op=linear_op,
prox_op=prox_op,
cost_op=cost_op,
lambda_init=1.0,
max_nb_of_iter=max_iter,
atol=1e-4,
verbose=1)
image_rec = pysap.Image(data=np.abs(x_final))
# image_rec.show()
recon_ssim = ssim(image_rec, image)
print('The Reconstruction SSIM is : ' + str(recon_ssim))
