-
Notifications
You must be signed in to change notification settings - Fork 0
/
3_model_vit_dual_attention_11_reparing.py
2096 lines (1667 loc) · 94.4 KB
/
3_model_vit_dual_attention_11_reparing.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
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# =============================================================================
# USER Guide
# =============================================================================
# in this file we interduce VIT and dual attention (position and channel)
# in segmentation of CMRI
# Notice : We use 3 output consist of 3 masks 'Right_Ventricle' , 'Myocard' ,'Left_Ventricle'
# and other data are assume as back ground
# load Libs: is loading lib in numerical and deep learning
# warrning off : is for switch GPU warning to off
# Initialize to Create : for Initialization train and test data
# modes: are modes for Net such as dual ,MSTGANET,NNUNET,CENET
# path_models='my models vit2' save folder
# path_name_save_every_epoch1 = '2_SBBVDAMLF_'
# path_name_save_every_epoch='path_name_save_every_epoch1'+name_back_bone+'_'+dual_attention_enable_model_N+'_loss '+str(loss_N)
# out_title : labels of 3 mask ['Right_Ventricle' , 'Myocard' ,'Left_Ventricle']
# function : our functions:
# color_mask
# find_index_good to find a good data for showing
# weight_loss : sum errors
# saver_result : save coloerd result
# show_and_save_coeff_exel in each epoch we save result in exel file
# Our ViT DUAL attention model
# class VIT_function(keras.layers.Layer): is base on ShiftedPatchTokenization
# model's functions to create our function on Vannil Unet
# def upsample_conv2d(img):
# def upsample(img):
# def downsample(img):
# def downsample2(img):
# init_layer(layer):
# blocks
# conv2d_block
# conv_block_simple
# conv_block_simple_no_bn
# identity_block
# BB_Resnet
# back_bone
# Encoder_Block0
# Encoder_Block
# Decoder_Block
# Decoder_Block0
# Decoder_Block1
# conv2d_block2
# VDAB_block
# DAB2in
# MSDAB
# VDAB_block
# DAB2in
# MSDAB
# BBVDAMLF_model
# see VIT model
# show and save all models if show_all_model is enable
# show model Dual attention if show_one_model is enable
# Training ablation study
# ablation_study
# we try print('FINISH 4 training stage 1 ')
# we try print('FINISH 7 training stage 2 ')
# we try Fine Tuning
# important saving
# name='final_model mode_'+mode+
# ' bb_'+ name_back_bone+
# ' DA_'+dual_attention_enable_model_N+' loss_N_'+str(loss_N)
# =============================================================================
# load Libs
# =============================================================================
# from models_RASTI_NN_Unet import get_model as get_model_NN_Unet
from models_RASTI_CE_Net import get_model as get_model_CE_Net
from models_RASTI_NN_Unet import nnUNet_2D as get_model_NN_Unet
import keras
from keras.models import Model
from tensorflow.keras.optimizers.legacy import Adam
import tensorflow as tf
from tensorflow.keras.layers.experimental.preprocessing import Resizing
from tensorflow.keras.layers import Reshape
from keras.layers import BatchNormalization , Activation, Dropout
# from keras.layers import BatchNormalization, Activation, Dropout
from keras.layers.convolutional import Conv2D, Conv2DTranspose
from keras.layers.pooling import MaxPooling2D
from keras.layers import Concatenate as concatenate
from keras.layers import UpSampling2D , AveragePooling2D
import matplotlib.pyplot as plt
import keras
from keras.utils.layer_utils import count_params
from math import isnan,isinf
import copy
from tensorflow.keras.constraints import max_norm
from tensorflow.keras.applications.resnet50 import ResNet50
# from tensorflow.keras.applications.
from tensorflow.keras.applications.inception_resnet_v2 import InceptionResNetV2
from tensorflow.keras.applications.mobilenet_v2 import MobileNetV2
from tensorflow.keras.applications.vgg16 import VGG16
from keras.layers.core import Activation, SpatialDropout2D
from DAttention import Channel_attention as CAM, Position_attention as PAM
from keras.layers import Input, BatchNormalization, Activation, Dense, Dropout
from keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau
from keras.models import Sequential
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator, array_to_img, img_to_array, load_img
import time
import datetime# import jdatetime# timee =jdatetime.date.today()
import metricss
import numpy as np
# from tensorflow.keras.applications.
import numpy as np
import pandas as pd
from saed import *
import keras as K
from skimage.transform import resize
import os
import time
time0 = time.time()
from metricss import *
# =============================================================================
# warrning off
# =============================================================================
import warnings
import sys
# import shutup; shutup.please()
if not sys.warnoptions:
warnings.simplefilter("ignore")
def fxn():warnings.warn("deprecated", DeprecationWarning)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
fxn()
warnings.filterwarnings("ignore")
warnings.warn('my warning')
with warnings.catch_warnings():
warnings.filterwarnings("ignore", category=DeprecationWarning)
warnings.filterwarnings("ignore", category=FutureWarning)
warnings.filterwarnings('ignore')
warnings.filterwarnings("ignore", message="divide by zero encountered in divide")
warnings.filterwarnings("ignore", message="divide by zero encountered")
warnings.filterwarnings("ignore", message="invalid value encountered")
plt.close('all')
# =============================================================================
# Initialize to Create
# =============================================================================
# modes=['dual']
# modes=['MSTGANET']
# modes=['NNUNET']
# modes=['CENET']
demo=False
demo=True
show_print=True
# show_print=False
path_models='my models vit2'
path_name_save_every_epoch1 = '2_SBBVDAMLF_'
img_size_target = 128
# detail_size_attention=True
detail_size_attention=False
show_one_model=False
# show_one_model=True
# porposed_model=True
# porposed_model=False
play_sound = True
play_sound = False
show_save_result_every_epochs=False
show_save_result_every_epochs =True
ablation_study=True
# ablation_study=False
# show_all_model=True
show_all_model=False
preprocessing=False
shutdown_after_Train_model=True
shutdown_after_Train_model=False
apply_our_losses=True
apply_our_losses=False
ceof_loss_callibrate=True
# ceof_loss_callibrate=False
global alfas4
global alfas7
global out_title
global loss_N
global loss_loss_loss
global loss1_cof1
global loss1_cof2
global loss1_cof3
global loss1_cof4
global loss1_cof5
global loss1_cof6
global loss1_cof7
global index001_train
global index001_val
global index001_test
out_title =['Right_Ventricle' , 'Myocard' ,'Left_Ventricle']
# =============================================================================
# TRAINING PARAMETERS
# =============================================================================
# metrics_all =['accuracy' ,'dice',tf.keras.metrics.MeanIoU(num_classes=2),tf.keras.metrics.MeanAbsoluteError,tf.keras.metrics.Precision,tf.keras.metrics.Recall,tf.keras.metrics.TrueNegatives,tf.keras.metrics.TrueNegatives(),tf.keras.metrics.FalseNegatives(),tf.keras.metrics.FalsePositives()]
# metrics_all =['accuracy' ,'dice',tf.keras.metrics.MeanIoU(num_classes=2), tf.keras.metrics.BinaryIoU(num_classes=2),\
# tf.keras.metrics.MeanAbsoluteError,tf.keras.metrics.Precision,tf.keras.metrics.Recall,tf.keras.metrics.TrueNegatives,tf.keras.metrics.TrueNegatives(),tf.keras.metrics.FalseNegatives(),tf.keras.metrics.FalsePositives()]
metrics_all =['accuracy' , dice_coef]
# modes=['CENET','NNUNET','dual','MSTGANET']
modes=['dual','MSTGANET','CENET','NNUNET']
# modes=[ 'MSTGANET' ,'dual']
dual_attention_enable_modelss=['vsc ','vs ','v c ',' sc ', 'v ',' c ',' s ','none',]
# dual_attention_enable_modelss=['vsc ']
loss_no=[1,4,7]
# loss_no=[7]
# loss_no=[1]
bb_enable_modelss=['resnet','none']
# bb_enable_modelss=['resnet' ]
epsilon = 10**-5
EPOCH7 = 1000
EPOCH4 = EPOCH7+200
EPOCH1 = copy.deepcopy(EPOCH4)
EPOCH_fine_tuning = 200
EPOCH_CAL=1
EPOCH7 = 200
EPOCH4 = EPOCH7+200
EPOCH1 = copy.deepcopy(EPOCH4)
EPOCH1=220
EPOCH_fine_tuning = 200
EPOCH_CAL=1
if demo:
EPOCH7 = 3
EPOCH4 = EPOCH7+2
EPOCH1 = copy.deepcopy(220)
EPOCH_fine_tuning = 2
EPOCH_CAL=1
INIT_LR = 1e-3
batch_size_no = 2
metricsss = [ 'accuracy',
# metricss.iou
]
# =============================================================================
# function
# =============================================================================
def color_mask(a,layer=1):
color_mask1=np.zeros( [np.shape(a)[0],np.shape(a)[1] ,3])
color_mask1[:,:,layer-1]=a[:,:,0]
return color_mask1
def find_index_good(Y_train10,Y_train20,Y_train30,th=0.1):
solid=[]
index001_train=0
for index001_train in range(len(Y_train10)):
tempp =np.mean(Y_train10[index001_train]) *np.mean(Y_train30[index001_train]) *np.mean(Y_train30[index001_train])
solid.append(tempp)
# if np.mean(Y_train10[index001_train]) >th and np.mean(Y_train20[index001_train]) >th and np.mean(Y_train30[index001_train]) >th :
# break
# # print(index001_train)
try:index001_train=np.where (solid ==np.max(solid))
except :s=1
try: index001_train=index001_train[0]
except :s=1
try: index001_train=index001_train[0]
except :s=1
try: index001_train=index001_train[0]
except :s=1
print('index001 = ' , index001_train)
return index001_train
def weight_loss (y_true, y_pred):
try:
import metricss
if loss_N ==4:
error=0
x1='l1=loss1_cof1 *' +loss_loss_loss[0] +'(y_true, y_pred)'
# print('123456798',x1)
# eval (x1)
eval ('l2=loss1_cof2 *' +loss_loss_loss[1] +'(y_true, y_pred)')
eval ('l3=loss1_cof3 *' +loss_loss_loss[2] +'(y_true, y_pred)')
eval ('l4=loss1_cof4 *' +loss_loss_loss[3] +'(y_true, y_pred)')
eval ('l5=loss1_cof5 *' +loss_loss_loss[4] +'(y_true, y_pred)')
# eval ('l6=loss1_cof6 *' +loss_loss_loss[5] +'(y_true, y_pred)')
# eval ('l7=loss1_cof7 *' +loss_loss_loss[6] +'(y_true, y_pred)')
error = l1+l2+l3+l4+l5
if loss_N ==7:
error=0
eval ('l1=loss1_cof1 *' +loss_loss_loss[0] +'(y_true, y_pred)')
eval ('l2=loss1_cof2 *' +loss_loss_loss[1] +'(y_true, y_pred)')
eval ('l3=loss1_cof3 *' +loss_loss_loss[2] +'(y_true, y_pred)')
eval ('l4=loss1_cof4 *' +loss_loss_loss[3] +'(y_true, y_pred)')
eval ('l5=loss1_cof5 *' +loss_loss_loss[4] +'(y_true, y_pred)')
eval ('l6=loss1_cof6 *' +loss_loss_loss[5] +'(y_true, y_pred)')
eval ('l7=loss1_cof7 *' +loss_loss_loss[6] +'(y_true, y_pred)')
error = l1+l2+l3+l4+l5+l6+l7
except:
from metricss import losss
error = losss(y_true, y_pred)
return error
def saver_result (model,X,Y1,Y2,Y3,index,path_name_save_every_epoch,current_epoch,
train_val_test='train'):
import matplotlib.pyplot as plt
import os
# index=4
# dicee=0.5
# print('ssss',X.shape)
p1,p2,p3 = model.predict(X , verbose=1)
# plt.figure();
XX,YY = masker (X,index,Y1,Y2,Y3)
XX1,YY1 = masker (X,index,p1,p2,p3)
dicee1 = np.mean (metricss.acc_coef(Y1[index], p1[index]))
dicee2 = np.mean (metricss.acc_coef(Y2[index], p2[index]))
dicee3 = np.mean (metricss.acc_coef(Y3[index], p3[index]))
diff1 = np.mean (np.power(Y1[index]- p1[index],2))
diff2 = np.mean (np.power(Y2[index]- p2[index],2))
diff3 = np.mean (np.power(Y3[index]- p3[index],2))
dicee1=100*np.round(dicee1,4)
dicee2=100*np.round(dicee2,4)
dicee3=100*np.round(dicee3,4)
diff1=100*np.round(diff1,4)
diff2=100*np.round(diff2,4)
diff3=100*np.round(diff3,4)
dicee=(dicee3+dicee2+dicee1)/3
difff=(diff3+diff2+diff1)/3
dicees=np.round(dicee,4)
diffs=np.round(difff,4)
plt.figure()
plt.subplot(141);plt.imshow(X[index]);plt.title('Original Image , index = '+str(index))
plt.subplot(142);plt.imshow(YY);plt.title('Ground Truth')
plt.subplot(143);plt.imshow(YY1);plt.title('Predicted '+str(dicees)+' %')
plt.subplot(143);plt.imshow(YY1);plt.title('Predicted MSE = '+str(diffs) )
plt.subplot(144);plt.imshow((YY-YY1));plt.title('Diffrence ')
try:os.mkdir(path_name_save_every_epoch)
except:print('hast')
os.startfile(path_name_save_every_epoch)
try:
# plt.legend();
figManager = plt.get_current_fig_manager()
figManager.window.showMaximized()
plt.pause(5);
plt.savefig ( path_name_save_every_epoch +'\\'+train_val_test+' F_index_'+str(index)+'ep = '+str(current_epoch)+'.png')
plt.close('all')
except:print('hast')
plt.figure()
dicee1s = np.round(dicee1,4)
dicee2s = np.round(dicee2,4)
dicee3s = np.round(dicee3,4)
diff1s = np.round(diff1,4)
diff2s = np.round(diff2,4)
diff3s = np.round(diff3,4)
cnt=0
plt.subplot(1,2,1);plt.imshow(X[index],cmap='gray');plt.title('Original Image , index = '+str(index))
cnt=4 ;plt.subplot(3,6,cnt);plt.imshow(Y1[index],cmap='gray');plt.title('Ground Truth '+out_title[0])
cnt=cnt+1;plt.subplot(3,6,cnt);plt.imshow(p1[index],cmap='gray');
temp1=color_mask (Y1[index] , 1)
temp2=color_mask (p1[index] , 1)
cnt=4 ;plt.subplot(3,6,cnt);plt.imshow(temp1,cmap='gray');plt.title('Ground Truth '+out_title[0])
cnt=cnt+1;plt.subplot(3,6,cnt);plt.imshow(temp2,cmap='gray');
plt.title('Predicted '+ str(dicee1s)+' %' )
plt.title('Predicted MSE = '+ str(diff1s) )
cnt=cnt+1;plt.subplot(3,6,cnt);plt.imshow(np.abs (Y1[index]-p1[index]),cmap='gray');plt.title('Diffrence ')
cnt=10 ;plt.subplot(3,6,cnt);plt.imshow(Y2[index],cmap='gray');plt.title('Ground Truth '+out_title[1])
cnt=cnt+1;plt.subplot(3,6,cnt);plt.imshow(p2[index],cmap='gray');
temp1=color_mask (Y2[index] , 2)
temp2=color_mask (p2[index] , 2)
cnt=10 ;plt.subplot(3,6,cnt);plt.imshow(temp1,cmap='gray');plt.title('Ground Truth '+out_title[0])
cnt=cnt+1;plt.subplot(3,6,cnt);plt.imshow(temp2,cmap='gray');
plt.title('Predicted '+ str(dicee2s)+' %' )
plt.title('Predicted MSE = '+ str(diff2s) )
cnt=cnt+1;plt.subplot(3,6,cnt);plt.imshow(np.abs (Y2[index]-p2[index]),cmap='gray');plt.title('Diffrence ')
cnt=16 ;plt.subplot(3,6,cnt);plt.imshow(Y3[index],cmap='gray');plt.title('Ground Truth '+out_title[2])
cnt=cnt+1;plt.subplot(3,6,cnt);plt.imshow(p3[index],cmap='gray');
temp1=color_mask (Y3[index] , 3)
temp2=color_mask (p3[index] , 3)
cnt=16 ;plt.subplot(3,6,cnt);plt.imshow(temp1,cmap='gray');plt.title('Ground Truth '+out_title[0])
cnt=cnt+1;plt.subplot(3,6,cnt);plt.imshow(temp2,cmap='gray');
plt.title('Predicted '+ str(dicee3s)+' %' )
plt.title('Predicted MSE = '+ str(diff3s) )
cnt=cnt+1;plt.subplot(3,6,cnt);plt.imshow(np.abs (Y3[index]-p3[index]),cmap='gray');plt.title('Diffrence ')
plt.subplot(1,2,1);plt.imshow(X[index],cmap='gray');plt.title ('Original Image , index = '+str(index))
# plt.subplot(144);plt.imshow(YY-YY1);plt.title('Diffrence ')
try:os.mkdir(path_name_save_every_epoch)
except:print('hast')
try:
figManager = plt.get_current_fig_manager()
figManager.window.showMaximized()
plt.pause(5);
plt.savefig ( path_name_save_every_epoch +'\\'+train_val_test+' index_'+str(index)+'ep = '+str(current_epoch)+'.png')
plt.close('all')
except:print('hast')
s=1
def show_and_save_coeff_exel ( epoch, model,history , file_exel ,show=True, save=True):
print (10*'*')
print ('show_and_save_coeff_exel')
print (10*'*')
final_metrics=[]
final_value=[]
try:
keyss=history.history.keys()
for key1 in keyss:
# if 'loss' in key1 or 'accuracy' in key1:
final_metrics.append(key1)
if show_print:
print('final_metrics',final_metrics)
for S in final_metrics:
v=history.history[S][-1]
final_value.append(v)
except:s=1
metrics1 = final_metrics
mvalue=final_value
sheet_name='model_parameters'
counter0=0;lenx=[];values=[];model_layers_names=[];indexes=[]
for i in range(np.shape(model.layers) [0]):
# print(i,model.layers[i].name)
if '_channel' in model.layers[i].name or '_position' in model.layers[i].name\
or '_VIT' in model.layers[i].name or '_original' in model.layers[i].name :
x=model.layers[i].name
xx = str(counter0)+' i='+str(i)+' layers.name = '+str(x)
lenx.append (8+len (xx))
for i in range(np.shape(model.layers) [0]):
if '_channel' in model.layers[i].name or '_position' in model.layers[i].name\
or '_VIT' in model.layers[i].name or '_original' in model.layers[i].name :
counter0=counter0+1
x=model.layers[i].name
v=tf.nn.softplus(model.layers[i].w).numpy()[0]
temp=''
if v>0:
temp=' '
t=''
if i<1000:t=' '
if i<100:t=' '
if i<10:t=' '
t1=''
if counter0<1000:t1=' '
if counter0<100:t1=' '
if counter0<10:t1=' '
xx = str(counter0)+t1+' i='+str(i)+t+' layers.name = '+str(x)
if show:
if show_print:
print (xx,(np.max(lenx)-len (xx))*' ','value = ',temp,v)
values.append(v)
model_layers_names.append(x)
indexes.append(counter0)
C=[]; V=[]
V.append(epoch)
C.append('epoch')
if save:
for m in mvalue:
V.append(m)
V.append('#')
for m in values:
V.append(m)
for m in metrics1:
C.append(m)
C.append('#')
for m in model_layers_names:
C.append(m)
df = pd.DataFrame([V], columns=C,)# index=indexes,
with pd.ExcelWriter(file_exel) as writer:
df.to_excel(writer, sheet_name=sheet_name)
return 1
# =============================================================================
# Keras costume layer
# =============================================================================
class coef_layer(keras.layers.Layer):
def __init__(self, name,**kwargs):
super(coef_layer, self).__init__(name=name)
w_init = tf.random_normal_initializer()
# w_init =tf.keras.constraints.NonNeg()
self.w = tf.Variable(
# initial_value =tf.keras.constraints.NonNeg(),
initial_value=w_init(shape=(1,), dtype="float32"),
trainable=True,
)
# self.name = name
#self.w = tf.nn.softplus(self.w2)
def call(self, inputs):
# tf.math.greater_equal(w,0.0)
# w.tf.cast(tf.gra
# self.add_loss(keras.constraints.non_neg(self.w))
# ln (e^x +1)
#self.w = tf.nn.softplus(self.w2)
return inputs * tf.nn.softplus(self.w)
# =============================================================================
# Our ViT DUAL attention model
# =============================================================================
# ShiftedPatchTokenization
class VIT_function(keras.layers.Layer):
def __init__(
self,
image_size=(None,128,128,1),
patch_size=32,
num_patches=128,
projection_dim=1280,
vanilla=True,
**kwargs,
):
super().__init__(**kwargs)
self.vanilla = vanilla # Flag to swtich to vanilla patch extractor
self.image_size = image_size
self.patch_size = patch_size
self.half_patch = patch_size // 2
self.flatten_patches = keras.layers.Reshape((num_patches, -1))
self.projection = keras.layers.Dense(units=projection_dim)
self.layer_norm = keras.layers.LayerNormalization(epsilon=1e-6)
def crop_shift_pad(self, images, mode):
# Build the diagonally shifted images
if mode == "left-up":
crop_height = self.half_patch
crop_width = self.half_patch
shift_height = 0
shift_width = 0
elif mode == "left-down":
crop_height = 0
crop_width = self.half_patch
shift_height = self.half_patch
shift_width = 0
elif mode == "right-up":
crop_height = self.half_patch
crop_width = 0
shift_height = 0
shift_width = self.half_patch
else:
crop_height = 0
crop_width = 0
shift_height = self.half_patch
shift_width = self.half_patch
# Crop the shifted images and pad them
crop = tf.image.crop_to_bounding_box(
images,
offset_height=crop_height,
offset_width=crop_width,
target_height=self.image_size - self.half_patch,
target_width=self.image_size - self.half_patch,
)
shift_pad = tf.image.pad_to_bounding_box(
crop,
offset_height=shift_height,
offset_width=shift_width,
target_height=self.image_size,
target_width=self.image_size,
)
return shift_pad
def get_config(self):
config = super().get_config()
config.update({
"image_size": self.image_size,
"patch_size": self.patch_size,
})
return config
def call(self, images):
if not self.vanilla:
# Concat the shifted images with the original image
images = tf.concat(
[
images,
self.crop_shift_pad(images, mode="left-up"),
self.crop_shift_pad(images, mode="left-down"),
self.crop_shift_pad(images, mode="right-up"),
self.crop_shift_pad(images, mode="right-down"),
],
axis=-1,
)
# Patchify the images and flatten it
patches = tf.image.extract_patches(
images=images,
sizes=[1, self.patch_size, self.patch_size, 1],
strides=[1, self.patch_size, self.patch_size, 1],
rates=[1, 1, 1, 1],
padding="VALID",
)
flat_patches = self.flatten_patches(patches)
if not self.vanilla:
# Layer normalize the flat patches and linearly project it
tokens = self.layer_norm(flat_patches)
tokens = self.projection(tokens)
else:
# Linearly project the flat patches
tokens = self.projection(flat_patches)
# return (tokens, patches)
tokens = keras.layers.Reshape(self.image_size[1:])(tokens)
tokens = tf.keras.activations.softmax(tokens)
#
return tokens
# myVit = VIT_function(image_size=(None,128,128,1),
# patch_size=32,
# num_patches=128,
# projection_dim=1280,
# vanilla=True,)
# print(myVit)
# stop_here_now
# =============================================================================
# end VIT
# =============================================================================
# =============================================================================
# model's functions
# =============================================================================
def upsample_conv2d(img):
if img_size_ori == img_size_target:
return img
return Conv2DTranspose(img, (img_size_target, img_size_target), mode='constant', preserve_range=True)
def upsample(img):
if img_size_ori == img_size_target:
return img
return resize(img, (img_size_target, img_size_target), mode='constant', preserve_range=True)
def downsample(img):
if img_size_ori == img_size_target:
return img
return resize(img, (img_size_ori, img_size_ori), mode='constant', preserve_range=True)
#return img[:img_size_ori, :img_size_ori]
def downsample2(img):
x = MaxPooling2D((2, 2))(img)
return x
# conv strid =2
def init_layer(layer):
session = K.get_session()
weights_initializer = tf.variables_initializer(layer.weights)
session.run(weights_initializer)
# =============================================================================
# blocks
# =============================================================================
def conv2d_block(input_tensor, n_filters=10, kernel_size = 3, batchnorm = True):
x = Conv2D(filters = n_filters, kernel_size = (kernel_size, kernel_size),\
kernel_initializer = 'he_normal', padding = 'same')(input_tensor)
x = BatchNormalization()(x)
x = Activation('relu')(x)
return x
def conv_block_simple(prevlayer, filters, prefix, strides=(1, 1)):
conv = Conv2D(filters, (3, 3), padding="same", kernel_initializer="he_normal", strides=strides, name=prefix + "_conv")(prevlayer)
conv = BatchNormalization(name=prefix + "_bn")(conv)
conv = Activation('relu', name=prefix + "_activation")(conv)
return conv
def conv_block_simple_no_bn(prevlayer, filters, prefix, strides=(1, 1)):
conv = Conv2D(filters, (3, 3), padding="same", kernel_initializer="he_normal", strides=strides, name=prefix + "_conv")(prevlayer)
conv = Activation('relu', name=prefix + "_activation")(conv)
return conv
def identity_block(input_tensor, kernel_size=3, filters=[3,3,3], stage='stage1', block='b1'):
filters1, filters2, filters3 = filters
if K.image_data_format() == 'channels_last':
bn_axis = 3
else:
bn_axis = 1
conv_name_base = 'res' + str(stage) + block + '_branch'
bn_name_base = 'bn' + str(stage) + block + '_branch'
x = Conv2D(filters1, (1, 1), name=conv_name_base + '2a')(input_tensor)
x = BatchNormalization(axis=bn_axis, name=bn_name_base + '2a')(x)
x = Activation('relu')(x)
x = Conv2D(filters2, kernel_size,
padding='same', name=conv_name_base + '2b')(x)
x = BatchNormalization(axis=bn_axis, name=bn_name_base + '2b')(x)
x = Activation('relu')(x)
x = Conv2D(filters3, (1, 1), name=conv_name_base + '2c')(x)
x = BatchNormalization(axis=bn_axis, name=bn_name_base + '2c')(x)
x = layers.add([x, input_tensor])
x = Activation('relu')(x)
return x
def BB_Resnet(w,h):
try:
base_model = ResNet50(include_top=False , input_shape=(128,128,3), pooling='avg', weights=None)
except:
# Transfer learning
# base_model=Keras.models.load ('ResNet50')
# base_model=Keras.models.load_weights('wresnet50')
s=1
# base_model = ResNet50(include_top=False, weights=None , input_shape=(w,h,3), pooling='avg')
"""
for i in range(143):
if base_model.layers[i].output.shape[1] == 64 and 'conv' in base_model.layers[i].name[-4:]:
i64=i;print(i, base_model.layers[i].name)
break
for i in range(143):
if base_model.layers[i].output.shape[1] == 32 and 'conv' in base_model.layers[i].name[-4:]:
i32=i;print(i, base_model.layers[i].name)
break
for i in range(143):
if base_model.layers[i].output.shape[1] == 16 and 'conv' in base_model.layers[i].name[-4:]:
i16=i;print(i, base_model.layers[i].name)
break
for i in range(143):
if base_model.layers[i].output.shape[1] == 8 and 'conv' in base_model.layers[i].name[-4:]:
i8=i;print(i, base_model.layers[i].name)
break
"""
resnet_base = keras.models.Model(base_model.input, base_model.layers[142].output)
# from keras.models import Model
input_shape=(w,h,3)
x=(w,h,3)
# Build model.
#model = Model(input_shape, x, name='resnet50')
# resnet_base = ResNet50(input_shape=input_shape, include_top=False)
for l in resnet_base.layers:
l.trainable = True
#conv1 = resnet_base.get_layer("input_7").output # ==> 128
conv1 = resnet_base.layers[0].output
#conv2 = resnet_base.get_layer("conv1_relu").output # Layer 4==> 64
conv2 = resnet_base.layers[4].output
conv3 = resnet_base.layers[38].output
conv4 = resnet_base.layers[80].output
conv5 = resnet_base.layers[142].output
# conv3 = resnet_base.get_layer("conlock3_out").output # Layer 38 ==> 32
# conv4 = resnet_base.get_layer("conv3_block4_out").output # Layer 80 ==> 16
# conv5 = resnet_base.get_layer("conv4_block6_out").output # Layer 142 ==> 8
up6 = concatenate(axis=-1)([UpSampling2D()(conv5), conv4])
conv6 = conv_block_simple(up6, 256, "conv6_1")
conv6 = conv_block_simple(conv6, 256, "conv6_2")
up7 = concatenate(axis=-1)([UpSampling2D()(conv6), conv3] )
conv7 = conv_block_simple(up7, 192, "conv7_1")
conv7 = conv_block_simple(conv7, 192, "conv7_2")
up8 = concatenate(axis=-1)([UpSampling2D()(conv7), conv2])
conv8 = conv_block_simple(up8, 128, "conv8_1")
conv8 = conv_block_simple(conv8, 128, "conv8_2")
up9 = concatenate(axis=-1)([UpSampling2D()(conv8), conv1])
conv9 = conv_block_simple(up9, 64, "conv9_1")
conv9 = conv_block_simple(conv9, 64, "conv9_2")
# up10 = UpSampling2D()(conv9)
conv10 = conv_block_simple(conv9, 32, "conv10_1")
conv10 = conv_block_simple(conv10, 32, "conv10_2")
conv10 = SpatialDropout2D(0.2)(conv10)
x = Conv2D(1, (1, 1), activation="sigmoid", name="prediction")(conv10)
model = Model(resnet_base.input, x)
return model
def back_bone(input_tensor,name ='resnet', BBtrainable=False):
# name_back_bone ='mobilenet_v2'
# name_back_bone ='vgg16'
# name_back_bone ='inception_resnet_v2
import numpy as np
# x = conv2d_block(input_tensor)
S = np.shape(input_tensor)
# print( 'input size = ',S)
im_height=S[1]; im_width=S[2]
from tensorflow.keras.applications.resnet50 import ResNet50
BB_model = BB_Resnet(S[0],S[1])
BB_model.trainable = BBtrainable
# print('BBmodel output: ', len(BB_model.layers), BB_model.output.shape)
# if 'resnet50' in name.lower():
# from tensorflow.keras.applications.resnet50 import ResNet50
# BB_model = ResNet50(include_top=False,input_shape=(im_height,im_width,1),weights=None,pooling='avg')
# print( 'ResNet50 is loaded')
# if 'mobile' in name.lower():
# from tensorflow.keras.applications.mobilenet_v2 import MobileNetV2
# BB_model = MobileNetV2(include_top=False,input_shape=(im_height,im_width,1),weights=None,pooling='avg')
# print( 'MobileNetV2 is loaded')
# if 'vgg' in name.lower():
# from tensorflow.keras.applications.vgg16 import VGG16
# BB_model = VGG16(include_top=False,input_shape=(im_height,im_width,1),weights=None,pooling='avg')
# print( 'VGG16 is loaded')
# if 'inception' in name.lower():
# from tensorflow.keras.applications.inception_resnet_v2 import InceptionResNetV2
# BB_model = InceptionResNetV2(include_top=False,input_shape=(im_height,im_width,1),weights=None,pooling='avg')
# print( 'InceptionResNetV2 is loaded')
f001 =BB_model (input_tensor)
# print (np.shape(f001))
a = BB_model.count_params()
# print('Total Parameters (Back BONE) = ',name, mil(a))
# =============================================================================
# shape and name layer
# =============================================================================
return f001
def Encoder_Block0(input_tensor, n_filters =10, kernel_size = 3, batchnorm = True,
dual_attention_enable_Encoder_Block0='sc',
section_name_Encoder_Block0 = 'section_name_Encoder_Block0'
):
x = conv2d_block(input_tensor, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x = conv2d_block(x, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x =VDAB_block(x ,3,
kernel_size = 3, batchnorm = True,
dual_attention_enable=dual_attention_enable_Encoder_Block0,
section_name = section_name_Encoder_Block0
)
return x
# Encoder_Block
# MAXPool ,Conv2DB ,Conv2DB
def Encoder_Block(input_tensor, n_filters=10, kernel_size = 3, batchnorm = True,
dual_attention_enable_Encoder_Block='sc',
section_name_Encoder_Block = 'section_name_Encoder_Block'
):
x = MaxPooling2D((2, 2))(input_tensor)
x = conv2d_block(x, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x = conv2d_block(x, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x =VDAB_block(x ,3,
kernel_size = 3, batchnorm = True,
dual_attention_enable=dual_attention_enable_Encoder_Block ,
section_name = section_name_Encoder_Block
)
return x
# Conv2DB ,Conv2DB
# Decoder_Block
# Conv2DB , Conv2DB Up-Sample, Conv2DB
def Decoder_Block(A2, B3, n_filters=10, kernel_size = 3, batchnorm = True,
dual_attention_enable_Decoder_Block='vsc',
section_name_Decoder_Block = 'section_name_Decoder_Block'
):
input_tensor = concatenate(axis=-1)([A2, B3] )
# x = MaxPooling2D((2, 2))(input_tensor)
x = conv2d_block(input_tensor, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x = conv2d_block(x, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x= UpSampling2D( size=(2, 2), interpolation="nearest")(x)
x = conv2d_block(x, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x =VDAB_block(x ,3,
kernel_size = 3, batchnorm = True,
dual_attention_enable=dual_attention_enable_Decoder_Block ,
section_name = section_name_Decoder_Block
)
return x
# Decoder_Block0
# Conv1 ×1, Sigmoid
def Decoder_Block0(A1,B2 , n_filters=10, kernel_size = 3, batchnorm = True,
dual_attention_enable_Decoder_Block0='vsc',
section_name_Decoder_Block0 = 'section_name_Decoder_Block0'
):
x = concatenate(axis=-1)([A1, B2])
D1 =VDAB_block(x ,n_filters,
kernel_size = 3, batchnorm = True,
dual_attention_enable=dual_attention_enable_Decoder_Block0 ,
section_name = section_name_Decoder_Block0
)
# x = conv2d_block(input_tensor, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x = Conv2D(1, (1, 1), activation='sigmoid')(D1)
# o1 = Conv2D(1, (1, 1), activation='sigmoid',name= 'R_V_')(x)
# o2 = Conv2D(1, (1, 1), activation='sigmoid',name= 'MyoCard')(x)
# o3 = Conv2D(1, (1, 1), activation='sigmoid',name= 'L_V_')(x)
o1 = Conv2D(1, (1, 1), activation='sigmoid',name= out_title[0])(x)
o2 = Conv2D(1, (1, 1), activation='sigmoid',name= out_title[1])(x)
o3 = Conv2D(1, (1, 1), activation='sigmoid',name= out_title[2])(x)
# out_title
return o1,o2,o3
# Backward Block1
# Up-Sample, Conv2DB
def Decoder_Block1(input_tensor, n_filters=10, kernel_size = 3, batchnorm = True
,
dual_attention_enable_Decoder_Block1='vsc',
section_name_Decoder_Block1 = 'section_name_Decoder_Block1'
):
# x = MaxPooling2D((2, 2))(input_tensor)
# x = conv2d_block(input_tensor, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
# x = conv2d_block(x, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x= UpSampling2D( size=(2, 2), interpolation="nearest")(input_tensor)
x = conv2d_block(x, n_filters * 1, kernel_size = 3, batchnorm = batchnorm)
x =VDAB_block(x ,n_filters,