conv_abs_lapl *= 2 / np.max(conv_abs_lapl)
conv_abs_lapl[:] = np.minimum(conv_abs_lapl, 1)
exponent = 2.0 - conv_abs_lapl
exponent.show(force_show=True)
# Disturb the exponent function
# var_exponent_arr[:, 100:150] = 2.0
# --- Set up the inverse problem --- #
# Assemble operators and functionals for the solver
gradient = odl.Gradient(reco_space, pad_mode='order1')
lin_ops = [odl.IdentityOperator(reco_space), gradient]
data_matching = odl.solvers.L2NormSquared(reco_space).translated(data)
varlp_func = variable_lp.VariableLpModular(gradient.range,
exponent,
impl='cython')
# Left-multiplication version
reg_param = 2e0
regularizer = reg_param * varlp_func
# Right-multiplication version
# reg_param = 8e-2
# regularizer = varlp_func * reg_param
g = [data_matching, regularizer]
f = odl.solvers.IndicatorBox(reco_space, 0, 255)
# Uncomment the combined callback to also display iterates
callback = (odl.solvers.CallbackShow('iterate', step=20, clim=[0, 255])
& odl.solvers.CallbackPrintIteration())
# callback = odl.solvers.CallbackPrintIteration()
post_conv = fourier.inverse * post_kernel_ft * fourier
conv_abs_lapl = np.maximum(post_conv(abs_lapl), 0)
conv_abs_lapl -= np.min(conv_abs_lapl)
conv_abs_lapl *= 2 / np.max(conv_abs_lapl)
conv_abs_lapl[:] = np.minimum(conv_abs_lapl, 1)
exponent = 2.0 - conv_abs_lapl
exponent.show(force_show=True)
# --- Set up the inverse problem --- #
# Assemble operators and functionals for the solver
gradient = odl.Gradient(reco_space, pad_mode='order1')
lin_ops = [odl.IdentityOperator(reco_space), gradient]
data_matching = odl.solvers.L2NormSquared(reco_space).translated(data)
varlp_func = variable_lp.VariableLpModular(gradient.range,
exponent,
impl='numba_parallel')
reg_param = 4.5e-3
regularizer = reg_param * varlp_func
g = [data_matching, regularizer]
f = odl.solvers.IndicatorBox(reco_space, 0, 1.0)
# Uncomment the combined callback to also display iterates
callback = (odl.solvers.CallbackShow('iterate', step=10, clim=[0, 1.0])
& odl.solvers.CallbackPrintIteration())
# callback = odl.solvers.CallbackPrintIteration()
# Solve with initial guess x = data.
# Step size parameters are selected to ensure convergence.
# See douglas_rachford_pd doc for more information.