def recons_pysap_3D(image, samples_loc, **kwargs):
linear_op = pyWavelet3(wavelet_name="bior6.8",
nb_scale=4)
fourier_op = NFFT3(samples=samples_loc, shape=image.shape)
kspace_obs = fourier_op.op(image)
gradient_op_cd = Gradient_pMRI(data=kspace_obs,
fourier_op=fourier_op)
x_final, transform, cost = condat_3D(
gradient_op=gradient_op_cd,
linear_op=linear_op,
**kwargs)
return x_final.data, cost
'name': 'Db4',
'sp3d_name': 'BiOrthogonalTransform3D',
'filter_id': 2
}, {
'name': 'Bior35',
'sp3d_name': 'BiOrthogonalTransform3D',
'filter_id': 12
}, {
'name': 'Bior44',
'sp3d_name': 'BiOrthogonalTransform3D',
'filter_id': 13
}]
obj_wt_pwt = [{
'name': 'pwt_' + ite,
'wt': pyWavelet3(wavelet_name=ite, nb_scale=nb_scale)
} for ite in list_pwt]
obj_wt_sp3d = [{
'name':
'sp3d_' + ite['name'],
'wt':
Wavelet2(wavelet_name=ite['sp3d_name'],
nb_scale=nb_scale - 1,
**{'type_of_filters': ite['filter_id']})
} for ite in list_sp3d]
# This is a list of wavelets, each item is a dic containing a key 'name' and a
# key 'wt' containing the object of the wt
obj_wt = obj_wt_pwt + obj_wt_sp3d
# in range(Il.shape[0])])
kspace_data = loadmat('/volatile/bsarthou/datas/XP_pysap/data_N320_sos_vds_nc1139/datavalues.mat')['datavalues']
kspace_data = np.moveaxis(kspace_data, -1, 0)
print(kspace_data.shape)
#############################################################################
# FISTA optimization
# ------------------
#
# We now want to refine the zero order solution using a FISTA optimization.
# Here no cost function is set, and the optimization will reach the
# maximum number of iterations. Fill free to play with this parameter.
# Start the FISTA reconstruction
max_iter = 200
linear_op = pyWavelet3(wavelet_name="sym8",
nb_scale=3)
# fourier_op = NFFT3(samples=samples,
# shape=(128, 128, 128))
fourier_op = NUFFT(samples=samples, shape=(152, 152, 104), platform='gpu')
print('Generate the zero order solution')
rec_0 = np.asarray([fourier_op.adj_op(kspace_data[l, :]) for l in range(32)])
imshow3D(np.squeeze(np.sqrt(np.sum(np.abs(rec_0)**2, axis=0))),
display=True)
gradient_op = Gradient_pMRI(data=kspace_data,
fourier_op=fourier_op,
linear_op=linear_op,
S=Smaps)
'name': 'Bior44',
'sp3d_name': 'BiOrthogonalTransform3D',
'filter_id': 13
}]
# Load input data
filename = '/volatile/bsarthou/datas/' \
'meas_MID14_gre_800um_iso_128x128x128_FID24.mat'
Iref = loadmat(filename)['ref']
for (i, pwt_name) in enumerate(list_pwt):
for (j, dic) in enumerate(list_sp3d):
print('Pywavelet:', pwt_name)
time_pwt_trans, time_pwt_recons, error_pwt = [], [], []
for k in range(R):
linear_op = pyWavelet3(wavelet_name=pwt_name, nb_scale=nb_scale)
start = time.clock()
coeffs = linear_op.op(Iref)
mid = time.clock()
recons1 = linear_op.adj_op(coeffs)
end = time.clock()
time_pwt_trans.append(mid - start)
time_pwt_recons.append(end - mid)
error_pwt.append(np.abs((Iref - recons1).mean()))
# print('Trans time:', mid - start, 'Recons time:', end - mid)
print('{}: Trans, recons, error=({},{}, {})'.format(
pwt_name, np.mean(time_pwt_trans), np.mean(time_pwt_recons),
np.mean(error_pwt)))
wt_type = dic['sp3d_name']