def _get_vars(self):
ndim = len(self.img_shape)
mps_ker_shape = [self.num_coils] + [self.mps_ker_width] * ndim
if self.coord is None:
img_ker_shape = [i + self.mps_ker_width -
1 for i in self.y.shape[1:]]
else:
grd_shape = sp.estimate_shape(self.coord)
img_ker_shape = [i + self.mps_ker_width - 1 for i in grd_shape]
self.img_ker = sp.dirac(
img_ker_shape, dtype=self.dtype, device=self.device)
with self.device:
self.mps_ker = self.device.xp.zeros(
mps_ker_shape, dtype=self.dtype)
admm_img = admm_app.run()
pl.ImagePlot(admm_img)
#%% md
## ADMM with circulant preconditioner
#%%
rho = 1
circ_precond = mr.circulant_precond(mps, coord=coord, device=device, lamda=rho)
img_shape = mps.shape[1:]
G = sp.linop.FiniteDifference(img_shape)
g = G.H * G * sp.dirac(img_shape)
g = sp.fft(g)
g = sp.to_device(g, device=device)
circ_precond = 1 / (1 / circ_precond + lamda * g)
img_shape = mps.shape[1:]
D = sp.linop.Multiply(img_shape, circ_precond)
P = sp.linop.IFFT(img_shape) * D * sp.linop.FFT(img_shape)
admm_cp_app = mr.app.TotalVariationRecon(
ksp, mps, lamda=lamda, coord=coord, max_iter=max_iter // max_cg_iter,
P=P, rho=rho,
solver='ADMM', max_cg_iter=max_cg_iter, device=device, save_objective_values=True)
admm_cp_img = admm_cp_app.run()
pl.ImagePlot(admm_cp_img)
