mmin = -0.3 # min data value
mmax = 2 # max data value
data[data < mmin] = mmin
data[data > mmax] = mmax
#data = (data-mmin)/(mmax-mmin)
# initial guess
u = np.zeros([det, det, det], dtype='float32')
lamd = np.zeros([ntheta, det, det], dtype='float32')
flow = np.zeros([ntheta, det, det, 2], dtype='float32')
psi = data.copy()
# optical flow parameters
pars = [0.5, 0, 256, 4, 5, 1.1, 4]
niter = 257
with lcg.SolverLam(det, det, det, det, ntheta, phi,
float(sys.argv[1])) as tslv:
with dc.SolverDeform(ntheta, det, det) as dslv:
rho = 0.5
h0 = psi
for k in range(niter):
# registration
tic()
flow = dslv.registration_flow_batch(psi,
data,
mmin,
mmax,
flow.copy(),
pars,
nproc=42)
t1 = toc()
tic()
# initial guess
u = np.zeros([det, det, det], dtype='float32')
psi1 = data.copy()
psi2 = np.zeros([3, det, det, det], dtype='float32')
lamd1 = np.zeros([ntheta, det, det], dtype='float32')
lamd2 = np.zeros([3, det, det, det], dtype='float32')
flow = np.zeros([ntheta, det, det, 2], dtype='float32')
# optical flow parameters
pars = [0.5, 1, 1024, 4, 5, 1.1, 4]
niter = 257
alpha = 8e-14
with lcg.SolverLam(det, det, det, det, ntheta, phi, 1e-2) as tslv:
with dc.SolverDeform(ntheta, det, det, 42) as dslv:
rho1 = 0.5
rho2 = 0.5
h01 = psi1
h02 = psi2
for k in range(niter):
# registration
tic()
flow = dslv.registration_flow_batch(psi1,
data,
mmin,
mmax,
F = np.ndarray.flatten(F)
res = us2eq(-x, F, 1e-3, N)
return res
n0 = 16
n1 = 16
n2 = 8
det = 8
ntheta = 8
phi = np.pi/3
theta = np.linspace(0,2*np.pi,ntheta,endpoint=False).astype('float32')
f = np.zeros([n0,n1,n2]).astype('complex64')
f[n0//8:3*n0//8,n1//4:3*n1//4,n2//4:3*n2//4]=1
with lcg.SolverLam(n0, n1, n2, det, ntheta, phi,1e-3) as slv:
g_gpu = slv.fwd_lam(f,theta)
g = fwd_laminography(f, theta, phi, det, [n0,n1,n2])
print('fwd accuracy:', np.linalg.norm(g_gpu-g)/np.linalg.norm(g_gpu))
ff = adj_laminography(g, theta, phi, det, [n0,n1,n2])
ff_gpu = slv.adj_lam(g_gpu, theta)
print('adj accuracy:', np.linalg.norm(ff_gpu-ff)/np.linalg.norm(ff_gpu))
print('Adj test 1')
print(np.sum(f*np.conj(ff)))
print(np.sum(g*np.conj(g)))
print('Adj test 2')
print(np.sum(f*np.conj(ff_gpu)))
print(np.sum(g_gpu*np.conj(g_gpu)))
gg_gpu = slv.fwd_lam(ff_gpu,theta)
print('Norm test')