-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathmap_tgv.py
362 lines (289 loc) · 13.7 KB
/
map_tgv.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
"""Compute TGV results for
[1] M. J. Ehrhardt, P. J. Markiewicz, and C.-B. Schoenlieb,
Faster PET reconstruction with non-smooth priors by randomization and
preconditioning, Phys. Med. Biol., 2019. 10.1088/1361-6560/ab3d07"""
from __future__ import print_function, division
import os
import numpy as np
folder_data_amyloid = '/home/me404/store/data/201611_PET_Pawel_amyloid'
folder_data_fdg = '/home/me404/store/data/201712_PET_Pawel_fdg'
folder_out = '/home/me404/store/projects/201804_SPDHG_PET/results'
folder_file = '/home/me404/store/repositories/gitbb_spdhg_pawel/python'
folder_odl = '/home/me404/store/repositories/github_myODL'
import sys
sys.path.append(folder_odl)
import misc
import mMR
from stochastic_primal_dual_hybrid_gradient import pdhg, spdhg
import odl
from odl.contrib import fom
from odl.solvers import CallbackPrintIteration, CallbackPrintTiming
#%% set parameters and create folder structure
filename = 'map_tgv'
nepoch = 30
nepoch_target = 5000
#nepoch_target = 500
datasets = ['fdg', 'amyloid10min']
datasets = ['fdg']
rho = 0.999 # algorithm parameter (rho < 1)
tol_step = 1e-6
folder_norms = '{}/norms'.format(folder_out)
misc.mkdir(folder_norms)
for dataset in datasets:
print('<<< ' + dataset)
if dataset == 'amyloid10min':
folder_data = folder_data_amyloid
planes = None
alphas = [4, 5]
betas = [0.5, 1]
data_suffix = 'rings0-64_span1_time3000-3600'
clim = [0, 1] # colour limits for plots
elif dataset == 'fdg':
folder_data = folder_data_fdg
planes = [85, 90, 46]
alphas = [2]
betas = [0.5]
data_suffix = 'rings0-64_span1'
clim = [0, 10] # colour limits for plots
folder_main = '{}/{}_{}'.format(folder_out, filename, dataset)
misc.mkdir(folder_main)
misc.mkdir('{}/py'.format(folder_main))
misc.mkdir('{}/logs'.format(folder_main))
for alpha, beta in [(a, b) for a in alphas for b in betas]:
print('<<< <<< alpha = {}, beta = {}'.format(alpha, beta))
folder_param = '{}/alpha{:.2g}_beta{:.2g}'.format(folder_main, alpha,
beta)
misc.mkdir(folder_param)
misc.mkdir('{}/pics'.format(folder_param))
folder_today = '{}/nepochs{}'.format(folder_param, nepoch)
misc.mkdir(folder_today)
misc.mkdir('{}/npy'.format(folder_today))
misc.mkdir('{}/pics'.format(folder_today))
misc.mkdir('{}/figs'.format(folder_today))
# load real data and convert to odl
file_data = '{}/data_{}.npy'.format(folder_data, data_suffix)
(data, background, factors, image, image_mr,
image_ct) = np.load(file_data)
Y = mMR.operator_mmr().range
data = Y.element(data)
background = Y.element(background)
factors = Y.element(factors)
# define operator
K = mMR.operator_mmr(factors=factors)
U = K.domain
norm_K = misc.norm(K, '{}/norm_1subset.npy'.format(folder_norms))
KL = misc.kullback_leibler(Y, data, background)
gradient = odl.Gradient(U)
Id = odl.IdentityOperator(gradient.range)
PD = [odl.PartialDerivative(U, i, method='backward',
pad_mode='symmetric')
for i in range(3)]
E = odl.operator.ProductSpaceOperator(
[[PD[0], 0, 0],
[0, PD[1], 0],
[0, 0, PD[2]],
[PD[1], PD[0], 0],
[PD[2], 0, PD[0]],
[0, PD[2], PD[1]]])
D = odl.ProductSpaceOperator([[gradient, -Id],
[0, E]])
norm_D = misc.norm(D, '{}/norm_D.npy'.format(folder_param))
norm_vfield = odl.PointwiseNorm(gradient.range)
def save_image(x, n, f):
misc.save_image(x[0].asarray(), n, f, planes=planes, clim=clim)
misc.save_image(norm_vfield(x[1]).asarray(), n + '_norm_vfield',
f, planes=planes)
c = float(norm_K) / float(norm_D)
D *= c
norm_D *= c
L1 = odl.solvers.SeparableSum(*[
(alpha / c) * odl.solvers.GroupL1Norm(gradient.range),
(alpha * beta / c) * odl.solvers.GroupL1Norm(E.range)])
g = odl.solvers.SeparableSum(
odl.solvers.IndicatorBox(U, lower=0),
odl.solvers.ZeroFunctional(gradient.range))
X = D.domain
P = [odl.ComponentProjection(X, i) for i in range(2)]
A_ = odl.BroadcastOperator(K * P[0], D)
f_ = odl.solvers.SeparableSum(KL, L1)
obj_fun = f_ * A_ + g # objective functional
fldr = '{}/pics'.format(folder_param)
if not os.path.exists('{}/gray_image_pet.png'.format(fldr)):
tmp = U.element()
tmp_op = mMR.operator_mmr()
tmp_op.toodl(image, tmp)
misc.save_image(tmp.asarray(), 'image_pet', fldr, planes=planes)
tmp_op.toodl(image_mr, tmp)
misc.save_image(tmp.asarray(), 'image_mr', fldr, planes=planes)
tmp_op.toodl(image_ct, tmp)
misc.save_image(tmp.asarray(), 'image_ct', fldr, planes=planes)
# --- get target --- BE CAREFUL, THIS TAKES TIME
file_target = '{}/target.npy'.format(folder_param)
if not os.path.exists(file_target):
print('file {} does not exist. Compute it.'.format(file_target))
A = A_
f = f_
norm_A = misc.norm(A, '{}/norm_tgv.npy'.format(folder_main))
sigma = rho / norm_A
tau = rho / norm_A
niter_target = nepoch_target
step = 10
cb = (CallbackPrintIteration(step=step, end=', ') &
CallbackPrintTiming(step=step, cumulative=False, end=', ') &
CallbackPrintTiming(step=step, cumulative=True,
fmt='total={:.3f} s'))
x_opt = X.zero()
odl.solvers.pdhg(x_opt, g, f, A, niter_target, tau, sigma,
callback=cb)
obj_opt = obj_fun(x_opt)
save_image(x_opt, 'target', '{}/pics'.format(folder_param))
np.save(file_target, (x_opt, obj_opt))
else:
print('file {} exists. Load it.'.format(file_target))
x_opt, obj_opt = np.load(file_target)
# define a function to compute statistic during the iterations
class CallbackStore(odl.solvers.Callback):
def __init__(self, alg, iter_save, iter_plot, niter_per_epoch):
self.iter_save = iter_save
self.iter_plot = iter_plot
self.iter_count = 0
self.alg = alg
self.out = []
self.niter_per_epoch = niter_per_epoch
def __call__(self, x, Kx=None, tmp=None, **kwargs):
if type(x) is list:
x = x[0]
k = self.iter_count
if k in self.iter_save:
obj = obj_fun(x)
psnr_opt = fom.psnr(x[0], x_opt[0])
diff_opt = (x[0] - x_opt[0]).norm() / x_opt[0].norm()
diff_opt_v = (x[1] - x_opt[1]).norm() / x_opt[1].norm()
self.out.append({'obj': obj, 'psnr_opt': psnr_opt,
'diff_opt': diff_opt,
'diff_opt_v': diff_opt_v})
if k in self.iter_plot:
save_image(x, '{}_{}'.format(self.alg,
int(k / niter_per_epoch)),
'{}/pics'.format(folder_today))
self.iter_count += 1
# set number of subsets for algorithms
nsub = {'PDHG1': 1, 'SPDHG2-21-bal': 21,
'SPDHG2-100-bal': 100, 'SPDHG2-252-bal': 252}
# %% run algorithms
algs = nsub.keys()
for alg in algs:
file_result = '{}/npy/{}.npy'.format(folder_today, alg)
if os.path.exists(file_result):
print('file {} does exist. Do NOT compute it.'
.format(file_result))
else:
print('file {} does not exist. Compute it.'
.format(file_result))
# define operator for subsets
if nsub[alg] > 1:
partition = mMR.partition_by_angle(nsub[alg])
tmp = mMR.operator_mmr(sino_partition=partition)
Ys = tmp.range
fctrs = Ys.element([factors[s, :] for s in partition])
d = Ys.element([data[s, :] for s in partition])
bg = Ys.element([background[s, :] for s in partition])
Ks = mMR.operator_mmr(factors=fctrs,
sino_partition=partition)
KLs = misc.kullback_leibler(Ys, d, bg)
norm_Ks = misc.norms(Ks, '{}/norm_{}subsets.npy'
.format(folder_norms, nsub[alg]))
list_K = [op * P[0] for op in Ks]
ops = list_K + [D]
A = odl.BroadcastOperator(*ops)
f = odl.solvers.SeparableSum(*(list(KLs.functionals) +
[L1]))
norm_Ai = list(norm_Ks) + [norm_D]
else:
A = A_
f = f_
norm_Ai = [norm_K, norm_D]
if alg.endswith('uni'):
prob = [1 / len(A)] * len(A)
elif alg.endswith('bal'):
prob = [0.5 / nsub[alg]] * nsub[alg] + [0.5]
elif alg.endswith('imp'):
prob = [nAi / sum(norm_Ai) for nAi in norm_Ai]
else:
prob = [1, 1]
niter_per_epoch = int(np.round(nsub[alg] / sum(prob[:-1])))
niter = nepoch * niter_per_epoch
iter_save, iter_plot = misc.what_to_save(niter_per_epoch,
nepoch)
# output function to be used with the iterations
step = 1
step = int(np.ceil(niter_per_epoch / 10))
cb = (CallbackPrintIteration(step=step, end=', ') &
CallbackPrintTiming(step=step, cumulative=False,
end=', ') &
CallbackPrintTiming(step=step, fmt='total={:.3f} s',
cumulative=True) &
CallbackStore(alg, iter_save, iter_plot,
niter_per_epoch))
x = X.zero() # initialise variable
cb(x)
if alg.startswith('PDHG1'):
norm_A = misc.norm(A, '{}/norm_tgv.npy'
.format(folder_main))
sigma = rho / norm_A
tau = rho / norm_A
pdhg(x, f, g, A, tau, sigma, niter, callback=cb)
elif alg.startswith('SPDHG1'):
sigma = [rho / nAi for nAi in norm_Ai]
tau = rho * min([pi / nAi
for pi, nAi in zip(prob, norm_Ai)])
spdhg(x, f, g, A, tau, sigma, niter, prob=prob,
callback=cb)
elif alg.startswith('PDHG2'):
one = A[0].domain.one()
tmp = A[0].range.element()
A[0](one, out=tmp)
tmp.ufuncs.maximum(tol_step, out=tmp)
sigma = [rho / tmp, rho / norm_D]
one = A[0].range.one()
tmp = A[0].domain.element()
A[0].adjoint(one, out=tmp)
tmp.ufuncs.maximum(tol_step, out=tmp)
tmp.ufuncs.maximum(norm_D, out=tmp)
tau = (1. / 2 * rho) / tmp
def fun_select(x):
return [1, 1]
spdhg(x, f, g, A, tau, sigma, niter, fun_select=fun_select,
callback=cb)
tau, sigma = [None, ] * 2
elif alg.startswith('SPDHG2'):
one = A.domain.one()
tmp = A.range.element()
A(one, out=tmp)
tmp.ufuncs.maximum(tol_step, out=tmp)
sigma = ([rho / t for t in tmp[:-1]] +
[rho / norm_D])
tmp = A.domain.element()
max_domain = tol_step * A.domain.one()
for pi, Ai in zip(prob[:-1], A[:-1]):
one = Ai.range.one()
Ai.adjoint(one, out=tmp)
tmp /= pi
tmp.ufuncs.maximum(max_domain, out=max_domain)
max_domain.ufuncs.maximum(norm_D / prob[-1],
out=max_domain)
tau = rho / max_domain
spdhg(x, f, g, A, tau, sigma, niter, prob=prob,
callback=cb)
else:
raise NameError('Algorithm not defined')
np.save(file_result, (iter_save, niter, niter_per_epoch, x,
cb.callbacks[1].out, nsub[alg], prob))
# %% show all methods
iter_save_v, out_v, niter_per_epoch_v = {}, {}, {}
for a in algs:
(iter_save_v[a], _, niter_per_epoch_v[a], _, out_v[a], _, _
) = np.load('{}/npy/{}.npy'.format(folder_today, a))
out = misc.resort_out(out_v, obj_opt)
misc.quick_visual_output(iter_save_v, algs, out, niter_per_epoch_v,
folder_today)