-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpgdips.py
622 lines (550 loc) · 34.2 KB
/
pgdips.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
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
import argparse
import csv
import math
import os.path
import time
from collections import namedtuple
import configparser
import PIL
import numpy as np
import torch
import torch.nn as nn
import skimage
from skimage.metrics import mean_squared_error
from net import skip
from net.losses import ExclusionLoss
from net.noise import get_noise
from utils.image_io import *
parser = argparse.ArgumentParser()
parser.add_argument('--config', dest='config', default='ihc.cfg', type=str, help='path of configuration file')
args = parser.parse_args()
class Separation(object):
def __init__(self):
# read configuration file and prepare input image
self.configfile = './config/' + args.config
self.config_name = self.configfile.split('/')[-1].split('.')[0]
self.process_config()
self.preprocess_image()
if not os.path.exists(self.output_folder):
os.makedirs(self.output_folder)
print(f"new directory is created at {self.output_folder}.")
# Initialize training parameters
self.loss_function = None
self.mses = []
self.total_loss = None
self.parameters = None
self.current_result = None
self.best_result = None
# Initialize stain deconvolution parameters
self.SCF1 = torch.tensor([1.0], device=torch.device("cuda:0"))
self.SCF2 = torch.tensor([1.0], device=torch.device("cuda:0"))
self.Con1_net_inputs = None
self.Con2_net_inputs = None
self.stain_net1_input = None
self.stain_net2_input = None
self.Con1_net = None
self.Con2_net = None
self.BG_net = None
self.Con1_out = None
self.Con2_out = None
self.BG_out = None
self.SeparationResult = namedtuple("SeparationResult", ['Stained1', 'Stained2', 'mse', 'SVec1', 'SVec2'])
if self.num_stains == 3:
self.SCF3 = torch.tensor([1.0], device=torch.device("cuda:0"))
self.Con3_net = None
self.Con3_net_inputs = None
self.stain_net3_input = None
self.Con3_out = None
self.SeparationResult = namedtuple("SeparationResult",
['Stained1', 'Stained2', 'Stained3', 'mse', 'SVec1', 'SVec2', 'SVec3'])
self._init_all()
def process_config(self):
self.config = configparser.ConfigParser()
self.config.read(self.configfile)
self.input = self.config['system']['input']
self.input_folder = self.config['system']['input_folder']
self.output_folder = self.config['system']['output_folder']
self.output_folder = os.path.join(self.output_folder, f'{self.config_name}')
self.input_depth = int(self.config['preprocess']['input_depth'])
self.input_size = int(self.config['preprocess']['input_size'])
self.roi_selection = self.config['preprocess']['roi_selection']
self.x_start = int(self.config['preprocess']['x_coor_start'])
self.y_start = int(self.config['preprocess']['y_coor_start'])
self.num_stains = int(self.config['stain']['num_stains'])
self.fixed_color = bool(self.config['stain']['fixed_color'])
self.default_stain_vector_1 = self.config['stain']['default_stain_vector_1']
self.default_stain_vector_1 = [float(x.strip('_')) for x in self.default_stain_vector_1.split(',')]
self.default_stain_vector_2 = self.config['stain']['default_stain_vector_2']
self.default_stain_vector_2 = [float(x.strip('_')) for x in self.default_stain_vector_2.split(',')]
self.use_SCF = bool(self.config['stain']['use_spectral_correction_factor'])
self.iters = int(self.config['network']['iterations'])
self.keep_color_iters = int(self.config['network']['keep_color_iterations'])
self.learning_rate = float(self.config['network']['init_learning_rate'])
self.plot_during_training = bool(self.config['network']['plot_during_training'])
self.show_every = int(self.config['network']['show_every'])
self.weight_l1 = float(self.config['loss']['weight_l1'])
self.weight_exclusion = float(self.config['loss']['weight_exclusion'])
self.weight_mse = float(self.config['loss']['weight_mse'])
if self.show_every > self.iters:
self.show_every = self.iters
def preprocess_image(self):
self.image_filename = self.input
self.image_path = os.path.join(self.input_folder, self.image_filename)
self.image_name = self.image_filename.split('.')[0]
# load() function has trouble loading some tiff images, replacing it with skimage.io.imread seems to work
# i = load(self.image_path)
i = skimage.io.imread(self.image_path)
i = Image.fromarray(i)
self.save_csv = os.path.join(self.output_folder, f'{self.config_name}_out.csv')
if self.roi_selection == 'resize':
i_crop = i.resize((self.input_size, self.input_size), resample=PIL.Image.BICUBIC)
elif self.roi_selection == 'centercrop':
center = [math.floor(i.height / 2), math.floor(i.width / 2)]
i_crop = i.crop((center[1] - self.input_size / 2, center[0] - self.input_size / 2,
center[1] + self.input_size / 2, center[0] + self.input_size / 2))
elif self.roi_selection == 'crop':
i_crop = i.crop((self.x_start, self.y_start, self.x_start + self.input_size, self.y_start + self.input_size))
self.save_csv = f'{self.config_name}_{self.x_start}_{self.y_start}.csv'
else:
i_crop = i
i_crop.show()
i_new = pil_to_np(i_crop)
i_new = RGB_to_OD(i_new)
self.image = i_new
################### Initialization functions ###################
def _init_all(self):
# Call initialization functions.
self._init_images()
self._init_nets()
self._init_inputs()
self._init_parameters()
self._init_losses()
def _init_images(self):
# Create augmented versions of input and convert them to torch tensors.
self.images = create_augmentations(self.image)
self.images_torch = [np_to_torch(image).type(torch.cuda.FloatTensor) for image in self.images]
save_image(self.image_name + "_original", self.images[0], self.output_folder)
def _init_inputs(self):
# Initialize input, default is using differnt random noise for different DIP modules
input_type = 'noise'
# input_type = 'meshgrid'
data_type = torch.cuda.FloatTensor
origin_noise = torch_to_np(get_noise(self.input_depth,
input_type,
(self.images_torch[0].shape[2],
self.images_torch[0].shape[3])).type(data_type).detach())
self.Con1_net_inputs = [np_to_torch(aug).type(data_type).detach() for aug in
create_augmentations(origin_noise)]
origin_noise = torch_to_np(get_noise(self.input_depth,
input_type,
(self.images_torch[0].shape[2],
self.images_torch[0].shape[3])).type(data_type).detach())
self.Con2_net_inputs = [np_to_torch(aug).type(data_type).detach() for aug in
create_augmentations(origin_noise)]
origin_noise = torch_to_np(get_noise(self.input_depth,
input_type,
(self.images_torch[0].shape[2],
self.images_torch[0].shape[3])).type(data_type).detach())
self.stain_net1_input = [np_to_torch(aug).type(data_type).detach() for aug in
create_augmentations(origin_noise)]
origin_noise = torch_to_np(get_noise(self.input_depth,
input_type,
(self.images_torch[0].shape[2],
self.images_torch[0].shape[3])).type(data_type).detach())
self.stain_net2_input = [np_to_torch(aug).type(data_type).detach() for aug in
create_augmentations(origin_noise)]
origin_noise = torch_to_np(get_noise(self.input_depth,
input_type,
(self.images_torch[0].shape[2],
self.images_torch[0].shape[3])).type(data_type).detach())
self.BG_net_inputs = [np_to_torch(aug).type(data_type).detach() for aug in
create_augmentations(origin_noise)]
if self.num_stains == 3:
origin_noise = torch_to_np(get_noise(self.input_depth,
input_type,
(self.images_torch[0].shape[2],
self.images_torch[0].shape[3])).type(data_type).detach())
self.Con3_net_inputs = [np_to_torch(aug).type(data_type).detach() for aug in
create_augmentations(origin_noise)]
origin_noise = torch_to_np(get_noise(self.input_depth,
input_type,
(self.images_torch[0].shape[2],
self.images_torch[0].shape[3])).type(data_type).detach())
self.stain_net3_input = [np_to_torch(aug).type(data_type).detach() for aug in
create_augmentations(origin_noise)]
def _init_parameters(self):
# Initialize gradients.
self.parameters = [p for p in self.Con1_net.parameters()] + \
[p for p in self.Con2_net.parameters()] + \
[p for p in self.BG_net.parameters()]
self.parameters += [p for p in self.stain1.parameters()]
self.parameters += [p for p in self.stain2.parameters()]
if self.use_SCF:
self.SCF1.requires_grad = True
self.SCF2.requires_grad = True
self.parameters += [self.SCF1, self.SCF2]
if self.num_stains == 3:
self.parameters += [p for p in self.Con3_net.parameters()]
self.parameters += [p for p in self.stain3.parameters()]
if self.use_SCF:
self.SCF3.requires_grad = True
self.parameters += [self.SCF3]
def _init_nets(self):
# Initialize network structures.
data_type = torch.cuda.FloatTensor
pad = 'reflection'
Con1_net = skip(
self.input_depth, 1,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.Con1_net = Con1_net.type(data_type)
Con2_net = skip(
self.input_depth, 1,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.Con2_net = Con2_net.type(data_type)
BG_net = skip(
self.input_depth, 3,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.BG_net = BG_net.type(data_type)
stain_net1 = skip(
self.input_depth, 3,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.stain1 = stain_net1.type(data_type)
stain_net2 = skip(
self.input_depth, 3,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.stain2 = stain_net2.type(data_type)
if self.num_stains == 3:
Con3_net = skip(
self.input_depth, 1,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.Con3_net = Con3_net.type(data_type)
stain_net3 = skip(
self.input_depth, 3,
num_channels_down=[8, 16, 32, 64, 128],
num_channels_up=[8, 16, 32, 64, 128],
num_channels_skip=[0, 0, 0, 4, 4],
upsample_mode='bilinear',
filter_size_down=5,
filter_size_up=5,
need_sigmoid=True, need_bias=True, pad=pad, act_fun='LeakyReLU')
self.stain3 = stain_net3.type(data_type)
def _init_losses(self):
# Initialize loss functions.
data_type = torch.cuda.FloatTensor
self.l1_loss = nn.L1Loss().type(data_type)
self.l1_loss_topk = nn.L1Loss(reduction='none').type(data_type)
self.mse_loss = nn.MSELoss().type(data_type)
self.exclusion_loss = ExclusionLoss().type(data_type)
################### Optimization functions ###################
def optimize(self):
# Optimize the network for one step.
torch.backends.cudnn.enabled = True
torch.backends.cudnn.benchmark = True
optimizer = torch.optim.Adam(self.parameters, lr=self.learning_rate)
for j in range(self.iters):
optimizer.zero_grad()
# train one step and obtain current results
self._optimization_closure(j)
self._obtain_current_result(j)
if self.plot_during_training:
self._plot_closure(j)
optimizer.step()
def get_stains(self):
# Normalize stain vectors.
self.current_stain1_normalized_0 = torch.divide(self.current_stain1[0][0], torch.sqrt(
torch.pow(self.current_stain1[0][0], 2) + torch.pow(self.current_stain1[0][1], 2) + torch.pow(
self.current_stain1[0][2], 2)))
self.current_stain1_normalized_1 = torch.divide(self.current_stain1[0][1], torch.sqrt(
torch.pow(self.current_stain1[0][0], 2) + torch.pow(self.current_stain1[0][1], 2) + torch.pow(
self.current_stain1[0][2], 2)))
self.current_stain1_normalized_2 = torch.divide(self.current_stain1[0][2], torch.sqrt(
torch.pow(self.current_stain1[0][0], 2) + torch.pow(self.current_stain1[0][1], 2) + torch.pow(
self.current_stain1[0][2], 2)))
self.current_stain2_normalized_0 = torch.divide(self.current_stain2[0][0], torch.sqrt(
torch.pow(self.current_stain2[0][0], 2) + torch.pow(self.current_stain2[0][1], 2) + torch.pow(
self.current_stain2[0][2], 2)))
self.current_stain2_normalized_1 = torch.divide(self.current_stain2[0][1], torch.sqrt(
torch.pow(self.current_stain2[0][0], 2) + torch.pow(self.current_stain2[0][1], 2) + torch.pow(
self.current_stain2[0][2], 2)))
self.current_stain2_normalized_2 = torch.divide(self.current_stain2[0][2], torch.sqrt(
torch.pow(self.current_stain2[0][0], 2) + torch.pow(self.current_stain2[0][1], 2) + torch.pow(
self.current_stain2[0][2], 2)))
if self.num_stains == 3:
self.current_stain3_normalized_0 = torch.divide(self.current_stain3[0][0], torch.sqrt(
torch.pow(self.current_stain3[0][0], 2) + torch.pow(self.current_stain3[0][1], 2) + torch.pow(
self.current_stain3[0][2], 2)))
self.current_stain3_normalized_1 = torch.divide(self.current_stain3[0][1], torch.sqrt(
torch.pow(self.current_stain3[0][0], 2) + torch.pow(self.current_stain3[0][1], 2) + torch.pow(
self.current_stain3[0][2], 2)))
self.current_stain3_normalized_2 = torch.divide(self.current_stain2[0][2], torch.sqrt(
torch.pow(self.current_stain3[0][0], 2) + torch.pow(self.current_stain3[0][1], 2) + torch.pow(
self.current_stain3[0][2], 2)))
def get_stained(self):
# Generate stained images
self.Stained1 = np.zeros((1, 3, self.input_size, self.input_size), dtype='float32')
self.Stained1[0, 0, :, :] = (self.Con1_out[0] * self.current_stain1_normalized_0).detach().cpu()
self.Stained1[0, 1, :, :] = (self.Con1_out[0] * self.current_stain1_normalized_1).detach().cpu()
self.Stained1[0, 2, :, :] = (self.Con1_out[0] * self.current_stain1_normalized_2).detach().cpu()
self.Stained2 = np.zeros((1, 3, self.input_size, self.input_size), dtype='float32')
self.Stained2[0, 0, :, :] = (self.Con2_out[0] * self.current_stain2_normalized_0).detach().cpu()
self.Stained2[0, 1, :, :] = (self.Con2_out[0] * self.current_stain2_normalized_1).detach().cpu()
self.Stained2[0, 2, :, :] = (self.Con2_out[0] * self.current_stain2_normalized_2).detach().cpu()
self.All = self.SCF1.item() * self.Stained1 + self.SCF2.item() * self.Stained2
if self.num_stains == 3:
self.Stained3 = np.zeros((1, 3, self.input_size, self.input_size), dtype='float32')
self.Stained3[0, 0, :, :] = (self.Con3_out[0] * self.current_stain3_normalized_0).detach().cpu()
self.Stained3[0, 1, :, :] = (self.Con3_out[0] * self.current_stain3_normalized_1).detach().cpu()
self.Stained3[0, 2, :, :] = (self.Con3_out[0] * self.current_stain3_normalized_2).detach().cpu()
self.All += self.SCF3.item() * self.Stained3
self.All_wb = np.zeros((1, 3, self.input_size, self.input_size), dtype='float32')
self.All_wb[0, 0, :, :] = self.All[0, 0, :, :] - torch.log10(self.current_BG[0][0]).item()
self.All_wb[0, 1, :, :] = self.All[0, 1, :, :] - torch.log10(self.current_BG[0][1]).item()
self.All_wb[0, 2, :, :] = self.All[0, 2, :, :] - torch.log10(self.current_BG[0][2]).item()
def _get_augmentation(self, iteration):
# Retrieve augmented images. Choose from 8 versions in #odd iters and use the original image in #even iters.
if iteration % 2 == 1:
return 0
iteration //= 2
return iteration % 8
def _optimization_closure(self, step):
# Calculate current stain deconvolution parameters and backpropagate gradients.
# get augmentations
if step == self.iters - 1:
reg_noise_std = 0
elif step < 1000:
reg_noise_std = (1 / 1000.) * (step // 100)
else:
reg_noise_std = 1 / 1000.
aug = self._get_augmentation(step)
if step == self.iters - 1:
aug = 0
# calculate parameters, including stain vectors, concentration maps, background correction factor, and SCFs
stain_net1_input = self.stain_net1_input[aug] + (self.stain_net1_input[aug].clone().normal_() * reg_noise_std)
self.current_stain1 = self.stain1(stain_net1_input)[:, :,
self.images_torch[aug].shape[2] // 2:self.images_torch[aug].shape[2] // 2 + 1,
self.images_torch[aug].shape[3] // 2:self.images_torch[aug].shape[
3] // 2 + 1] * 0.9 + 0.05
stain_net2_input = self.stain_net2_input[aug] + (self.stain_net2_input[aug].clone().normal_() * reg_noise_std)
self.current_stain2 = self.stain2(stain_net2_input)[:, :,
self.images_torch[aug].shape[2] // 2:self.images_torch[aug].shape[2] // 2 + 1,
self.images_torch[aug].shape[3] // 2:self.images_torch[aug].shape[
3] // 2 + 1] * 0.9 + 0.05
BG_net_input = self.BG_net_inputs[aug] + (self.BG_net_inputs[aug].clone().normal_() * reg_noise_std)
self.current_BG = self.BG_net(BG_net_input)[:, :,
self.images_torch[aug].shape[2] // 2:self.images_torch[aug].shape[2] // 2 + 1,
self.images_torch[aug].shape[3] // 2:self.images_torch[aug].shape[3] // 2 + 1] * 0.9 + 0.05
Con1_net_input = self.Con1_net_inputs[aug] + (
self.Con1_net_inputs[aug].clone().normal_() * reg_noise_std)
Con2_net_input = self.Con2_net_inputs[aug] + (
self.Con2_net_inputs[aug].clone().normal_() * reg_noise_std)
self.Con1_out = self.Con1_net(Con1_net_input)
self.Con2_out = self.Con2_net(Con2_net_input)
if self.num_stains == 3:
Con3_net_input = self.Con3_net_inputs[aug] + (
self.Con3_net_inputs[aug].clone().normal_() * reg_noise_std)
self.Con3_out = self.Con3_net(Con3_net_input)
stain_net3_input = self.stain_net3_input[aug] + (
self.stain_net3_input[aug].clone().normal_() * reg_noise_std)
self.current_stain3 = self.stain3(stain_net3_input)[:, :,
self.images_torch[aug].shape[2] // 2:self.images_torch[aug].shape[2] // 2 + 1,
self.images_torch[aug].shape[3] // 2:self.images_torch[aug].shape[
3] // 2 + 1] * 0.9 + 0.05
# get normalized stain vectors and generate stained images
self.get_stains()
self.get_stained()
self.Con1R = self.SCF1 * self.Con1_out[0] * self.current_stain1_normalized_0
self.Con1G = self.SCF1 * self.Con1_out[0] * self.current_stain1_normalized_1
self.Con1B = self.SCF1 * self.Con1_out[0] * self.current_stain1_normalized_2
self.Con2R = self.SCF2 * self.Con2_out[0] * self.current_stain2_normalized_0
self.Con2G = self.SCF2 * self.Con2_out[0] * self.current_stain2_normalized_1
self.Con2B = self.SCF2 * self.Con2_out[0] * self.current_stain2_normalized_2
if self.num_stains == 2:
self.Rnorm = self.Con1R + self.Con2R - torch.log10(self.current_BG[0][0])
self.Gnorm = self.Con1G + self.Con2G - torch.log10(self.current_BG[0][1])
self.Bnorm = self.Con1B + self.Con2B - torch.log10(self.current_BG[0][2])
else:
self.Con3R = self.SCF3 * self.Con3_out[0] * self.current_stain3_normalized_0
self.Con3G = self.SCF3 * self.Con3_out[0] * self.current_stain3_normalized_1
self.Con3B = self.SCF3 * self.Con3_out[0] * self.current_stain3_normalized_2
self.Rnorm = self.Con1R + self.Con2R + self.Con3R - torch.log10(self.current_BG[0][0])
self.Gnorm = self.Con1G + self.Con2G + self.Con3G - torch.log10(self.current_BG[0][1])
self.Bnorm = self.Con1B + self.Con2B + self.Con3B - torch.log10(self.current_BG[0][2])
self.OD = self.images_torch[aug][0][0] + self.images_torch[aug][0][1] + self.images_torch[aug][0][2]
self.OD_pred = torch.squeeze(self.Rnorm + self.Gnorm + self.Bnorm)
self.total_loss = self.weight_l1 * self.l1_loss(torch.squeeze(self.Rnorm), self.images_torch[aug][0][0])
self.total_loss += self.weight_l1 * self.l1_loss(torch.squeeze(self.Gnorm), self.images_torch[aug][0][1])
self.total_loss += self.weight_l1 * self.l1_loss(torch.squeeze(self.Bnorm), self.images_torch[aug][0][2])
self.total_loss += self.weight_l1 * self.l1_loss(self.OD, self.OD_pred)
self.total_loss += self.weight_exclusion * self.exclusion_loss(self.Con1_out.repeat(1, 3, 1, 1), self.Con2_out.repeat(1, 3, 1, 1))
if self.num_stains == 3:
self.total_loss += self.weight_exclusion * self.exclusion_loss(self.Con1_out.repeat(1, 3, 1, 1),
self.Con3_out.repeat(1, 3, 1, 1))
self.total_loss += self.weight_exclusion * self.exclusion_loss(self.Con2_out.repeat(1, 3, 1, 1),
self.Con3_out.repeat(1, 3, 1, 1))
self.total_loss -= 0.01 * self.l1_loss(self.current_stain1, self.current_stain2)
self.total_loss -= 0.01 * self.l1_loss(self.current_stain1, self.current_stain3)
self.total_loss -= 0.01 * self.l1_loss(self.current_stain2, self.current_stain3)
if step < self.keep_color_iters:
self.total_loss += self.weight_mse * self.mse_loss(self.current_stain1,
torch.tensor([[[[self.default_stain_vector_1[0]]], [[self.default_stain_vector_1[1]]], [[self.default_stain_vector_1[2]]]]]).type(
torch.cuda.FloatTensor))
self.total_loss += self.weight_mse * self.mse_loss(self.current_stain2,
torch.tensor([[[[self.default_stain_vector_2[0]]], [[self.default_stain_vector_2[1]]], [[self.default_stain_vector_2[2]]]]]).type(
torch.cuda.FloatTensor))
self.total_loss += self.weight_mse * self.mse_loss(self.current_BG,
torch.tensor([[[[1.]], [[1.]], [[1.]]]]).type(torch.cuda.FloatTensor))
self.total_loss += self.weight_mse * self.mse_loss(self.SCF1, torch.tensor([1.]).type(torch.cuda.FloatTensor))
self.total_loss += self.weight_mse * self.mse_loss(self.SCF2, torch.tensor([1.]).type(torch.cuda.FloatTensor))
if self.num_stains == 3:
self.total_loss += self.weight_mse * self.mse_loss(self.SCF3, torch.tensor([1.]).type(torch.cuda.FloatTensor))
self.total_loss.backward()
################### Output&Visualization functions ###################
def _obtain_current_result(self, step):
"""
puts in self.current result the current result.
also updates the best result
:return:
"""
if step == self.iters - 1 or step % 50 == 0 or step % self.show_every == self.show_every - 1:
Con1_out_np = np.clip(torch_to_np(self.Con1_out), 0, 1)
Con2_out_np = np.clip(torch_to_np(self.Con2_out), 0, 1)
image_out_np = np.zeros((1, 3, self.input_size, self.input_size), dtype='float32')
if self.num_stains == 2:
image_out_np[0, 0, :, :] = (self.Con1_out[0] * self.current_stain1_normalized_0 + self.Con2_out[0]
* self.current_stain2_normalized_0 - torch.log10(
self.current_BG[0][0])).detach().cpu()
image_out_np[0, 1, :, :] = (self.Con1_out[0] * self.current_stain1_normalized_1 + self.Con2_out[0]
* self.current_stain2_normalized_1 - torch.log10(
self.current_BG[0][1])).detach().cpu()
image_out_np[0, 2, :, :] = (self.Con1_out[0] * self.current_stain1_normalized_2 + self.Con2_out[0]
* self.current_stain2_normalized_2 - torch.log10(
self.current_BG[0][2])).detach().cpu()
else:
image_out_np[0, 0, :, :] = (self.Con1_out[0] * self.current_stain1_normalized_0 + self.Con2_out[0]
* self.current_stain2_normalized_0 + self.Con3_out[0]
* self.current_stain3_normalized_0 - torch.log10(
self.current_BG[0][0])).detach().cpu()
image_out_np[0, 1, :, :] = (self.Con1_out[0] * self.current_stain1_normalized_1 + self.Con2_out[0]
* self.current_stain2_normalized_1 + self.Con3_out[0]
* self.current_stain3_normalized_1 - torch.log10(
self.current_BG[0][1])).detach().cpu()
image_out_np[0, 2, :, :] = (self.Con1_out[0] * self.current_stain1_normalized_2 + self.Con2_out[0]
* self.current_stain2_normalized_2 + self.Con3_out[0]
* self.current_stain3_normalized_2 - torch.log10(
self.current_BG[0][2])).detach().cpu()
mse = mean_squared_error(self.images[0], image_out_np[0])
self.mses.append(mse)
file = open(self.save_csv, 'a+', newline='')
if self.num_stains == 2:
self.current_result = self.SeparationResult(Stained1=Con1_out_np, Stained2=Con2_out_np,
mse=mse, SVec1=self.current_stain1, SVec2=self.current_stain2)
with file:
# identifying header
header = ['Step', 'S1R', 'S1G', 'S1B', 'S2R', 'S2G', 'S2B', 'SCF1', 'SCF2', 'BGR', 'BGG', 'BGB']
writer = csv.DictWriter(file, fieldnames=header)
# writing data row-wise into the csv file
writer.writeheader()
writer.writerow({'Step': step,
'S1R': self.current_stain1_normalized_0.item(),
'S1G': self.current_stain1_normalized_1.item(),
'S1B': self.current_stain1_normalized_2.item(),
'S2R': self.current_stain2_normalized_0.item(),
'S2G': self.current_stain2_normalized_1.item(),
'S2B': self.current_stain2_normalized_2.item(),
'SCF1': self.SCF1.item(), 'SCF2': self.SCF2.item(),
'BGR': self.current_BG[0][0].item(),
'BGG': self.current_BG[0][1].item(),
'BGB': self.current_BG[0][2].item()})
else:
Con3_out_np = np.clip(torch_to_np(self.Con3_out), 0, 1)
self.current_result = self.SeparationResult(Stained1=Con1_out_np, Stained2=Con2_out_np, Stained3=Con3_out_np,
mse=mse, SVec1=self.current_stain1, SVec2=self.current_stain2,
SVec3=self.current_stain3)
with file:
# identifying header
header = ['Step', 'S1R', 'S1G', 'S1B', 'S2R', 'S2G', 'S2B', 'S3R', 'S3G', 'S3B', 'SCF1', 'SCF2',
'SCF3', 'BGR', 'BGG', 'BGB']
writer = csv.DictWriter(file, fieldnames=header)
# writing data row-wise into the csv file
writer.writeheader()
writer.writerow({'Step': step,
'S1R': self.current_stain1_normalized_0.item(),
'S1G': self.current_stain1_normalized_1.item(),
'S1B': self.current_stain1_normalized_2.item(),
'S2R': self.current_stain2_normalized_0.item(),
'S2G': self.current_stain2_normalized_1.item(),
'S2B': self.current_stain2_normalized_2.item(),
'S3R': self.current_stain3_normalized_0.item(),
'S3G': self.current_stain3_normalized_1.item(),
'S3B': self.current_stain3_normalized_2.item(),
'SCF1': self.SCF1.item(), 'SCF2': self.SCF2.item(), 'SCF3': self.SCF3.item(),
'BGR': self.current_BG[0][0].item(),
'BGG': self.current_BG[0][1].item(),
'BGB': self.current_BG[0][2].item()})
if self.best_result is None or self.best_result.mse < self.current_result.mse:
self.best_result = self.current_result
def _plot_closure(self, step):
print('Iteration {:5d} Loss {:5f} MSE_gt: {:f}'.format(step,
self.total_loss.item(),
self.current_result.mse),
'\n', end='')
if step % self.show_every == self.show_every - 1:
self.save_results(step)
def finalize(self):
save_graph(self.image_name + "_mse", self.mses, self.output_folder)
def save_results(self, step):
save_image(self.image_name + f"_Con1_{step}", self.current_result.Stained1, self.output_folder)
save_image(self.image_name + f"_Con2_{step}", self.current_result.Stained2, self.output_folder)
save_image(self.image_name + f"_Stained1_{step}", self.Stained1[0], self.output_folder)
save_image(self.image_name + f"_Stained2_{step}", self.Stained2[0], self.output_folder)
save_image(self.image_name + f"_WB_{step}", self.All_wb[0], self.output_folder)
save_image(self.image_name + f"_All_{step}", self.All[0], self.output_folder)
if self.num_stains == 3:
save_image(self.image_name + f"_Con3_{step}", self.current_result.Stained3, self.output_folder)
save_image(self.image_name + f"_Stained3_{step}", self.Stained3[0], self.output_folder)
print(f"Step: {step}, SVec1 and SVec2 values are:")
print(self.SCF1)
print(self.SCF2)
if __name__ == "__main__":
t_start = time.time()
s = Separation()
s.optimize()
t_end = time.time()
print(f"--- {int((t_end - t_start) / 60)} minutes and {(t_end - t_start) % 60} seconds ---")
s.finalize()