-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
1155 lines (813 loc) · 57.5 KB
/
main.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
import os
import re
import time
import math
import sys
import network
import shutil
import logging
import numpy as np
import losses_helper
import tensorflow as tf
from six.moves import xrange
import tensorflow.contrib.slim as slim
from tensorflow.python import debug as tf_debug
from tensorflow.python.client import device_lib
from datetime import datetime
# import ijremote
import configparser as configp
import data_reader
'''
TRAINING:
driving = 200
flying = 22390
monkaa = 6050
ptb = 20332
--------------
48972
TESTING:
driving = 100
flying = 4370
monkaa = 2614
ptb = 1210
--------------
8294
'''
class DatasetReader:
def delete_files_in_directories(self,dir_path):
fileList = os.listdir(dir_path)
for fileName in fileList:
os.remove(dir_path+"/"+fileName)
def create_and_remove_directories(self,dir_path,clean_existing_files,load_from_ckpt,testing_enabled):
if not os.path.exists(dir_path):
os.makedirs(dir_path+'/train')
os.makedirs(dir_path+'/test')
else:
if clean_existing_files == True and load_from_ckpt == False and testing_enabled == False:
self.delete_files_in_directories(dir_path+'/train')
self.delete_files_in_directories(dir_path+'/test')
def create_input_pipeline(self,sections):
prefix = self.FLAGS['DATASET_FOLDER']
self.dataset_used = 3
# memory_folder = '/dev/shm/'
self.filenames_train = ['driving_TRAIN.tfrecords','flying_TRAIN.tfrecords','monkaa_TRAIN.tfrecords','ptb_TRAIN.tfrecords','mid_TEST.tfrecords']
self.filenames_test = ['driving_TEST.tfrecords','flying_TEST.tfrecords','monkaa_TEST.tfrecords','ptb_TEST.tfrecords']
# driving
if sections[self.section_type] == sections[0]:
self.log()
self.log('Using Driving Dataset ... ')
self.log()
train_filenames = [prefix+self.filenames_train[1]]
test_filenames = [prefix+self.filenames_test[1]]
self.dataset_used = 2
# driving, flying
elif sections[self.section_type] == sections[1]:
self.log()
self.log('Using Flying, Monkaa Datasets ... ')
self.log()
train_filenames = [prefix+self.filenames_train[1],prefix+self.filenames_train[2]]
test_filenames = [prefix+self.filenames_test[1],prefix+self.filenames_test[2]]
self.dataset_used = 3
# driving, flying, monkaa
elif sections[self.section_type] == sections[2]:
self.log()
self.log('Using Flying, Monkaa and PTB Datasets ... ')
self.log()
train_filenames = [prefix+self.filenames_train[1],prefix+self.filenames_train[2],prefix+'/ptb/'+self.filenames_train[3]]
test_filenames = [prefix+self.filenames_test[1],prefix+self.filenames_test[2],prefix+'/ptb/'+self.filenames_test[3]]
self.dataset_used = 4
# driving, flying, monkaa, ptb
elif sections[self.section_type] == sections[3]:
self.log()
self.log('Using Driving, Flying, Monkaa and PTB Datasets ... ')
self.log()
train_filenames = [prefix+self.filenames_train[0],prefix+self.filenames_train[1],prefix+self.filenames_train[2],prefix+self.filenames_train[3],prefix+self.filenames_train[4]]
test_filenames = [prefix+self.filenames_test[0],prefix+self.filenames_test[1],prefix+self.filenames_test[2],prefix+self.filenames_test[3],prefix+self.filenames_train[4]]
self.dataset_used = 5
# only testing with ptb
elif sections[self.section_type] == sections[4]:
self.log()
self.log('Using PTB Dataset ... ')
self.log()
train_filenames = [prefix+self.filenames_train[3]]
test_filenames = [prefix+self.filenames_test[3]]
self.dataset_used = 1
train_dataset = data_reader.read_with_dataset_api(self.FLAGS['BATCH_SIZE'],train_filenames,version='1')
test_dataset = data_reader.read_with_dataset_api(self.FLAGS['TEST_BATCH_SIZE'],test_filenames,version='2')
train_iterator = train_dataset.make_initializable_iterator()
test_iterator = test_dataset.make_initializable_iterator()
return train_iterator, test_iterator
def preprocess(self):
file = './configs/training.ini'
self.section_type = 2
parser = configp.ConfigParser()
parser.read(file)
sections = parser.sections()
self.FLAGS = {
# TRAIN
'BATCH_SIZE': int(parser[sections[self.section_type]]['BATCH_SIZE']),
'TRAIN_DIR': parser[sections[self.section_type]]['TRAIN_DIR'],
'LOAD_FROM_CKPT': parser[sections[self.section_type]].getboolean('LOAD_FROM_CKPT'),
'DEBUG_MODE': parser[sections[self.section_type]].getboolean('DEBUG_MODE'),
'TOWER_NAME': parser[sections[self.section_type]]['TOWER_NAME'],
'MAX_STEPS': int(parser[sections[self.section_type]]['MAX_STEPS']),
'LOG_DEVICE_PLACEMENT': parser[sections[self.section_type]].getboolean('LOG_DEVICE_PLACEMENT'),
'NUM_EPOCHS_PER_DECAY': int(parser[sections[self.section_type]]['NUM_EPOCHS_PER_DECAY']),
'SHUFFLE_BATCH_QUEUE_CAPACITY': int(parser[sections[self.section_type]]['SHUFFLE_BATCH_QUEUE_CAPACITY']),
'SHUFFLE_BATCH_THREADS': int(parser[sections[self.section_type]]['SHUFFLE_BATCH_THREADS']),
'SHUFFLE_BATCH_MIN_AFTER_DEQUEUE': int(parser[sections[self.section_type]]['SHUFFLE_BATCH_MIN_AFTER_DEQUEUE']),
'NUM_GPUS': int(parser[sections[self.section_type]]['NUM_GPUS']),
'MOVING_AVERAGE_DECAY': float(parser[sections[self.section_type]]['MOVING_AVERAGE_DECAY']),
'TOTAL_TRAIN_EXAMPLES': int(parser[sections[self.section_type]]['TOTAL_TRAIN_EXAMPLES']),
'CLEAN_FILES': parser[sections[self.section_type]].getboolean('CLEAN_FILES'),
'TRAIN_WITH_PTB' : parser[sections[self.section_type]].getboolean('TRAIN_WITH_PTB'),
'DATASET_FOLDER': parser[sections[self.section_type]]['DATASET_FOLDER'],
'LOAD_WITH_NEW_LEARNING_RATE':parser[sections[self.section_type]].getboolean('LOAD_WITH_NEW_LEARNING_RATE'),
# TEST
'TESTING_ENABLED': parser[sections[self.section_type]].getboolean('TESTING_ENABLED'),
'TOTAL_TEST_EXAMPLES': int(parser[sections[self.section_type]]['TOTAL_TEST_EXAMPLES']),
'TEST_BATCH_SIZE': int(parser[sections[self.section_type]]['TEST_BATCH_SIZE']),
'TEST_AFTER_EPOCHS': int(parser[sections[self.section_type]]['TEST_AFTER_EPOCHS']),
'TOTAL_TRAIN_EXAMPLES': int(parser[sections[self.section_type]]['TOTAL_TRAIN_EXAMPLES']),
'TEST_ON_PTB_ONLY': parser[sections[self.section_type]].getboolean('TEST_ON_PTB_ONLY'),
# LR
'START_LEARNING_RATE': float(parser[sections[self.section_type]]['START_LEARNING_RATE']),
'END_LEARNING_RATE': float(parser[sections[self.section_type]]['END_LEARNING_RATE']),
'POWER': int(parser[sections[self.section_type]]['POWER'])
}
self.create_and_remove_directories(self.FLAGS['TRAIN_DIR'],
self.FLAGS['CLEAN_FILES'],
self.FLAGS['LOAD_FROM_CKPT'],
self.FLAGS['TESTING_ENABLED'])
self.TRAIN_DIR_LIST = self.FLAGS['TRAIN_DIR'].split('/')
# gives the # of steps required to complete 1 epoch
self.TRAIN_EPOCH = math.ceil(self.FLAGS['TOTAL_TRAIN_EXAMPLES'] / self.FLAGS['BATCH_SIZE'])
self.TEST_EPOCH = math.ceil(self.FLAGS['TOTAL_TEST_EXAMPLES'] / self.FLAGS['TEST_BATCH_SIZE'])
train_iterator, test_iterator = self.create_input_pipeline(sections)
# for testing
self.X = tf.placeholder(dtype=tf.float32, shape=(self.FLAGS['TEST_BATCH_SIZE'] * self.dataset_used, 224, 384, 8))
self.Y = tf.placeholder(dtype=tf.float32, shape=(self.FLAGS['TEST_BATCH_SIZE'] * self.dataset_used, 224, 384, 3))
return train_iterator, test_iterator
def train(self,iterator_train,iterator_test):
self.global_step = tf.get_variable(
'global_step', [],
initializer=tf.constant_initializer(0), trainable=False)
decay_steps = self.FLAGS['MAX_STEPS']
start_learning_rate = self.FLAGS['START_LEARNING_RATE']
end_learning_rate = self.FLAGS['END_LEARNING_RATE']
power = self.FLAGS['POWER']
#lr_new_gstep = tf.placeholder(tf.int32,name='global_stepo')
# lr_new_gstep = tf.get_variable('global_step', [], dtype=tf.int64, initializer=tf.constant_initializer(0), trainable=False)
# self.global_step = lr_new_gstep
self.alternate_global_step = tf.placeholder(tf.int32)
# self.incr_global_step = tf.assign(self.alternate_global_step, self.alternate_global_step+1)
# learning_rate = tf.placeholder(tf.float32,shape=(),name="learing_rate")
learning_rate = tf.train.polynomial_decay(start_learning_rate, self.alternate_global_step,
decay_steps, end_learning_rate,
power=power,name="new_one2")
opt = tf.train.AdamOptimizer(learning_rate)
# images, labels = tf.train.shuffle_batch(
# [ features_train['input_n'] , features_train['label_n'] ],
# batch_size=self.FLAGS['BATCH_SIZE'],
# capacity=self.FLAGS['SHUFFLE_BATCH_QUEUE_CAPACITY'],
# num_threads=self.FLAGS['SHUFFLE_BATCH_THREADS'],
# min_after_dequeue=self.FLAGS['SHUFFLE_BATCH_MIN_AFTER_DEQUEUE'],
# enqueue_many=False)
# batch_queue = tf.contrib.slim.prefetch_queue.prefetch_queue(
# [images, labels], capacity=self.FLAGS['SHUFFLE_BATCH_QUEUE_CAPACITY'] * self.FLAGS['NUM_GPUS'])
# if self.FLAGS['TESTING_ENABLED'] == True:
# self.images_test, self.labels_test = tf.train.shuffle_batch(
# [ features_test['input_n'] , features_test['label_n'] ],
# batch_size=self.FLAGS['TEST_BATCH_SIZE'],
# capacity=self.FLAGS['SHUFFLE_BATCH_QUEUE_CAPACITY'],
# num_threads=self.FLAGS['SHUFFLE_BATCH_THREADS'],
# min_after_dequeue=self.FLAGS['SHUFFLE_BATCH_MIN_AFTER_DEQUEUE'],
# enqueue_many=False)
# self.batch_queue_test = tf.contrib.slim.prefetch_queue.prefetch_queue(
# [self.images_test, self.labels_test], capacity=self.FLAGS['SHUFFLE_BATCH_QUEUE_CAPACITY'] * self.FLAGS['NUM_GPUS'])
tower_grads = []
train_summaries = []
test_summaries = []
with tf.variable_scope(tf.get_variable_scope()):
for i in xrange(self.FLAGS['NUM_GPUS']):
with tf.device('/gpu:%d' % i):
with tf.name_scope('%s_%d' % ('tower', i)) as scope:
# Dequeues one batch for the GPU
image_batch, label_batch, files1, files2 = self.combine_batches_from_datasets(iterator_train.get_next())
image_batch_test, label_batch_test, files1_test, files2_test = self.combine_batches_from_datasets(iterator_test.get_next())
# Calculate the loss for one tower of the CIFAR model. This function
# constructs the entire CIFAR model but shares the variables across
# all towers.
self.loss = self.tower_loss(scope, image_batch, label_batch)
_ = self.tower_loss(scope, image_batch_test, label_batch_test,'_test')
# Reuse variables for the next tower.
tf.get_variable_scope().reuse_variables()
# Retain the summaries from the final tower.
for item in tf.get_collection(tf.GraphKeys.SUMMARIES, scope):
if '_test' in item.name:
test_summaries.append(item)
else:
train_summaries.append(item)
# Calculate the gradients for the batch of data on this CIFAR tower.
grads = opt.compute_gradients(self.loss)
# grads = [(tf.clip_by_value(grad, -1., 1.), var) for grad, var in grads]
# Keep track of the gradients across all towers.
tower_grads.append(grads)
# We must calculate the mean of each gradient. Note that this is the
# synchronization point across all towers.
grads = self.average_gradients(tower_grads)
if not self.section_type is 4:
# Add a summary to track the learning rate.
train_summaries.append(tf.summary.scalar('learning_rate', learning_rate))
# Add histograms for gradients.
for grad, var in grads:
if grad is not None:
train_summaries.append(tf.summary.histogram(var.op.name + '/gradients', grad))
# Apply the gradients to adjust the shared variables.
apply_gradient_op = opt.apply_gradients(grads, global_step=self.global_step)
# Add histograms for trainable variables.
for var in tf.trainable_variables():
train_summaries.append(tf.summary.histogram(var.op.name, var))
# Track the moving averages of all trainable variables.
variable_averages = tf.train.ExponentialMovingAverage(
self.FLAGS['MOVING_AVERAGE_DECAY'], self.global_step)
variables_averages_op = variable_averages.apply(tf.trainable_variables())
# Group all updates to into a single train op.
train_op = tf.group(apply_gradient_op, variables_averages_op)
# Build the summary operation from the last tower summaries.
self.summary_op = tf.summary.merge(train_summaries)
self.summary_op_test = tf.summary.merge(test_summaries)
all_vars = tf.global_variables()
evolution1_vars = []
for i in all_vars:
if not 's_evolution2' in i.op.name :
evolution1_vars.append(i)
saver = tf.train.Saver(evolution1_vars)
saver2 = tf.train.Saver(tf.global_variables())
# Build an initialization operation to run below.
init = tf.global_variables_initializer()
# Start running operations on the Graph. allow_soft_placement must be set to
# True to build towers on GPU, as some of the ops do not have GPU
# implementations.
tf_config = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=self.FLAGS['LOG_DEVICE_PLACEMENT'])
if 'dacky' in os.uname()[1]:
logging.info('Dacky: Running with memory usage limits')
# change tf_config for dacky to use only 1 GPU
tf_config.gpu_options.per_process_gpu_memory_fraction = 0.6
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
else:
# change tf_config for lmb_cluster so that GPU is visible and utilized
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
sess = tf.Session(config=tf_config)
sess.run(init)
sess.run(train_iterator.initializer)
sess.run(iterator_test.initializer)
if self.FLAGS['LOAD_FROM_CKPT'] == True or self.section_type == 4:
print('loading from ckpt...')
saver.restore(sess,tf.train.latest_checkpoint(self.FLAGS['TRAIN_DIR']+'/train/'))
# saver.restore(sess,tf.train.latest_checkpoint('./ckpt/driving/one_at_a_time_training_flying/train/'))
# Start the queue runners.
tf.train.start_queue_runners(sess=sess)
# for debugging
if self.section_type == 4:
if os.path.exists(self.FLAGS['TRAIN_DIR'] +'/ptb_test'):
shutil.rmtree(self.FLAGS['TRAIN_DIR'] +'/ptb_test')
os.makedirs(self.FLAGS['TRAIN_DIR'] +'/ptb_test')
self.test_summary_writer = tf.summary.FileWriter(self.FLAGS['TRAIN_DIR']+'/ptb_test', sess.graph)
else:
summary_writer = tf.summary.FileWriter(self.FLAGS['TRAIN_DIR']+'/train', sess.graph)
self.test_summary_writer = tf.summary.FileWriter(self.FLAGS['TRAIN_DIR']+'/test', sess.graph)
# just to make sure we start from where we left, if load_from_ckpt = True
loop_start = tf.train.global_step(sess, self.global_step)
loop_stop = loop_start + self.FLAGS['MAX_STEPS']
if self.FLAGS['DEBUG_MODE']:
sess = tf_debug.LocalCLIDebugWrapperSession(sess)
first_iteration = True
# this will print in console for which time we are calculating the test loss.
# first time or second time or third time and more
test_loss_calculating_index = 1
# this will not train and only test the model on ptb dataset.
if self.section_type == 4:
self.test_model_on_ptb(sess,iterator_test)
sys.exit('Done Testing')
alternate_global_stepper = 0
# main loop
for step in range(loop_start,loop_stop):
start_time = time.time()
# lr_gstep = mozi
# print(type(lr_gstep))
# fetches = {"global_step": self.global_step, "incr_global_step": self.incr_global_step}
_, loss_value = sess.run([train_op, self.loss], feed_dict={
self.alternate_global_step: alternate_global_stepper
})
files_r1, files_r2 = sess.run([files1, files2])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
# ####################### Testing #######################
# # after every 10 epochs. calculate test loss
# if step % (self.TRAIN_EPOCH * FLAGS.TEST_AFTER_EPOCHS) == 0 and first_iteration==False:
# message = 'Printing Test loss for '+str(test_loss_calculating_index)+' time'
# self.log()
# self.log(message)
# self.log()
# self.perform_testing(sess,step)
# # increment index to know how many times we've calculated the test loss
# test_loss_calculating_index = test_loss_calculating_index + 1
# ####################### Testing #######################
if step % 100 == 0 or first_iteration==True:
num_examples_per_step = self.FLAGS['BATCH_SIZE'] * self.FLAGS['NUM_GPUS']
examples_per_sec = num_examples_per_step / duration
sec_per_batch = duration / self.FLAGS['NUM_GPUS']
first_iteration = False
format_str = ('%s: step %d, DIR: %s, loss = %.15f (%.1f examples/sec; %.3f '
'sec/batch)')
self.log(message=(format_str % (datetime.now(),step,self.TRAIN_DIR_LIST[-1], loss_value,
examples_per_sec, sec_per_batch)))
# if loss_value > 100:
# summary_str = sess.run(self.summary_op)
# summary_writer.add_summary(summary_str, step)
if step % 100 == 0 and step!=0:
summary_str = sess.run(self.summary_op, feed_dict={
self.alternate_global_step: alternate_global_stepper
})
summary_writer.add_summary(summary_str, step)
# # write a summary for test
# test_image_batch, test_label_batch = sess.run([test_image_batch, test_label_batch])
# loss_value, summary_str_test = sess.run([self.loss,self.summary_op_test],feed_dict={
# self.X: test_image_batch,
# self.Y: test_label_batch
# })
summary_str = sess.run(self.summary_op_test)
self.test_summary_writer.add_summary(summary_str, step)
alternate_global_stepper += 1
# Save the model checkpoint periodically.
if step % 5000 == 0 or (step + 1) == self.FLAGS['MAX_STEPS']:
checkpoint_path = os.path.join(self.FLAGS['TRAIN_DIR']+'/train', 'model.ckpt')
saver2.save(sess, checkpoint_path, global_step=step)
# if step == self.FLAGS['MAX_STEPS']+2:
# break
summary_writer.close()
self.test_summary_writer.close()
def test_model_on_ptb(self,sess,iterator_test):
self.log()
self.log(message='Testing ..., Total Test Epochs = ' + str(self.TEST_EPOCH))
self.log()
# iterator_test = sess.run(iterator_test)
for step in range(0,self.TEST_EPOCH + 10):
image_batch, label_batch = self.combine_batches_from_datasets(iterator_test.get_next())
image_batch, label_batch = self.get_network_input_forward(image_batch,label_batch)
image,label = sess.run([image_batch, label_batch])
summary_str = sess.run([self.summary_op],feed_dict={self.X: image})
format_str = ('%s: Testing step %d, loss = %.15f')
self.log(message=(format_str % (datetime.now(), step, np.log10(loss_value))))
# if step % 50 == 0:
self.test_summary_writer.add_summary(summary_str, step)
self.log()
self.log(message='Continue Training ...')
self.log()
def combine_batches_from_datasets(self,batches):
imgs = []
lbls = []
file1 = []
file2 = []
# driving
# batches[x][y] = (4, 2, 224, 384, 8)
imgs.append(batches[0][0])
lbls.append(batches[0][1])
file1.append(batches[0][2])
file2.append(batches[0][3])
if self.section_type > 0 and self.section_type is not 4:
imgs.append(batches[1][0])
lbls.append(batches[1][1])
file1.append(batches[1][2])
file2.append(batches[1][3])
if self.section_type > 1 and self.section_type is not 4:
imgs.append(batches[2][0])
lbls.append(batches[2][1])
file1.append(batches[2][2])
file2.append(batches[2][3])
if self.section_type > 2 and self.section_type is not 4:
imgs.append(batches[3][0])
lbls.append(batches[3][1])
file1.append(batches[3][2])
file2.append(batches[3][3])
# imgs.append(batches[3][0])
# lbls.append(batches[3][1])
final_img_batch = tf.concat(tuple(imgs),axis=0)
final_lbl_batch = tf.concat(tuple(lbls),axis=0)
final_f1_batch = tf.concat(tuple(file1),axis=0)
final_f2_batch = tf.concat(tuple(file2),axis=0)
return final_img_batch, final_lbl_batch,final_f1_batch ,final_f2_batch
# warped_img = lhpl.flow_warp(img2_to_tensor,pred_flow_to_tensor)
# warped_img = sess.run(warped_img)
# warped_img = np.squeeze(warped_img)
# # Image.fromarray(np.uint8(img2_orig)).show()
# Image.fromarray(np.uint8(warped_img)).show()
# print(loss)
def further_resize_imgs_lbls(self,network_input_images,network_input_labels):
network_input_images = tf.image.resize_images(network_input_images,[160,256],method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)
network_input_labels = tf.image.resize_images(network_input_labels,[160,256],method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)
network_input_labels_u = network_input_labels[:,:,:,0] * 0.714285714
network_input_labels_v = network_input_labels[:,:,:,1] * 0.666666667
network_input_labels_w = network_input_labels[:,:,:,2]
network_input_labels_u = tf.expand_dims(network_input_labels_u,axis=-1)
network_input_labels_v = tf.expand_dims(network_input_labels_v,axis=-1)
network_input_labels_w = tf.expand_dims(network_input_labels_w,axis=-1)
network_input_labels = tf.concat([network_input_labels_u,network_input_labels_v,network_input_labels_w],axis=3)
return network_input_images, network_input_labels
def remove_ptb_records(self,network_input_images,network_input_labels):
return network_input_images[0:8,:,:,:], network_input_labels[0:8,:,:,:]
def remove_mid_records(self,network_input_images,network_input_labels):
if self.section_type == 3:
return network_input_images[0:16,:,:,:], network_input_labels[0:16,:,:,:]
elif self.section_type == 2:
return network_input_images[0:12,:,:,:], network_input_labels[0:12,:,:,:]
elif self.section_type == 1:
return network_input_images[0:8,:,:,:], network_input_labels[0:8,:,:,:]
else:
return network_input_images[0:4,:,:,:], network_input_labels[0:4,:,:,:]
def write_forward_backward_images(self,forward,backward,forward_flow,backward_flow,summary_type='_train'):
# driving
forward_input_images = tf.concat([forward[0:4,:,:,0:3],forward[0:4,:,:,4:7]],axis=-2)
forward_input_depths = tf.concat([tf.expand_dims(forward[0:4,:,:,3],axis=-1),tf.expand_dims(forward[0:4,:,:,7],axis=-1)],axis=-2)
backward_input_images = tf.concat([backward[0:4,:,:,0:3],backward[0:4,:,:,4:7]],axis=-2)
backward_input_depths = tf.concat([tf.expand_dims(backward[0:4,:,:,3],axis=-1),tf.expand_dims(backward[0:4,:,:,7],axis=-1)],axis=-2)
tf.summary.image('input_images_forward_flying'+summary_type,forward_input_images)
tf.summary.image('input_depths_forward_flying'+summary_type,forward_input_depths)
tf.summary.image('input_images_backward_flying'+summary_type,backward_input_images)
tf.summary.image('input_depths_backward_flying'+summary_type,backward_input_depths)
if self.section_type > 0 and self.section_type is not 4:
# flying
forward_input_images = tf.concat([forward[4:8,:,:,0:3],forward[4:8,:,:,4:7]],axis=-2)
forward_input_depths = tf.concat([tf.expand_dims(forward[4:8,:,:,3],axis=-1),tf.expand_dims(forward[4:8,:,:,7],axis=-1)],axis=-2)
backward_input_images = tf.concat([backward[4:8,:,:,0:3],backward[4:8,:,:,4:7]],axis=-2)
backward_input_depths = tf.concat([tf.expand_dims(backward[4:8,:,:,3],axis=-1),tf.expand_dims(backward[4:8,:,:,7],axis=-1)],axis=-2)
tf.summary.image('input_images_forward_monkaa'+summary_type,forward_input_images)
tf.summary.image('input_depths_forward_monkaa'+summary_type,forward_input_depths)
tf.summary.image('input_images_backward_monkaa'+summary_type,backward_input_images)
tf.summary.image('input_depths_backward_monkaa'+summary_type,backward_input_depths)
if self.section_type > 1 and self.section_type is not 4:
# monkaa
forward_input_images = tf.concat([forward[8:12,:,:,0:3],forward[8:12,:,:,4:7]],axis=-2)
forward_input_depths = tf.concat([tf.expand_dims(forward[8:12,:,:,3],axis=-1),tf.expand_dims(forward[8:12,:,:,7],axis=-1)],axis=-2)
backward_input_images = tf.concat([backward[8:12,:,:,0:3],backward[8:12,:,:,4:7]],axis=-2)
backward_input_depths = tf.concat([tf.expand_dims(backward[8:12,:,:,3],axis=-1),tf.expand_dims(backward[8:12,:,:,7],axis=-1)],axis=-2)
tf.summary.image('input_images_forward_ptb'+summary_type,forward_input_images)
tf.summary.image('input_depths_forward_ptb'+summary_type,forward_input_depths)
tf.summary.image('input_images_backward_ptb'+summary_type,backward_input_images)
tf.summary.image('input_depths_backward_ptb'+summary_type,backward_input_depths)
# ground truth
label_flow_u = tf.concat([tf.expand_dims(forward_flow[0:4,:,:,0],axis=-1),tf.expand_dims(backward_flow[0:4,:,:,0],axis=-1)],axis=-2)
label_flow_v = tf.concat([tf.expand_dims(forward_flow[0:4,:,:,1],axis=-1),tf.expand_dims(backward_flow[0:4,:,:,1],axis=-1)],axis=-2)
tf.summary.image('label_flow_u_driving'+summary_type,label_flow_u)
tf.summary.image('label_flow_v_driving'+summary_type,label_flow_v)
if self.section_type > 0 and self.section_type is not 4:
label_flow_u = tf.concat([tf.expand_dims(forward_flow[4:8,:,:,0],axis=-1),tf.expand_dims(backward_flow[4:8,:,:,0],axis=-1)],axis=-2)
label_flow_v = tf.concat([tf.expand_dims(forward_flow[4:8,:,:,1],axis=-1),tf.expand_dims(backward_flow[4:8,:,:,1],axis=-1)],axis=-2)
tf.summary.image('label_flow_u_flying'+summary_type,label_flow_u)
tf.summary.image('label_flow_v_flying'+summary_type,label_flow_v)
if self.section_type > 1 and self.section_type is not 4:
label_flow_u = tf.concat([tf.expand_dims(forward_flow[8:12,:,:,0],axis=-1),tf.expand_dims(backward_flow[8:12,:,:,0],axis=-1)],axis=-2)
label_flow_v = tf.concat([tf.expand_dims(forward_flow[8:12,:,:,1],axis=-1),tf.expand_dims(backward_flow[8:12,:,:,1],axis=-1)],axis=-2)
tf.summary.image('label_flow_u_monkaa'+summary_type,label_flow_u)
tf.summary.image('label_flow_v_monkaa'+summary_type,label_flow_v)
if self.section_type > 2 and self.section_type is not 4:
forward_input_images = tf.concat([forward[12:16,:,:,0:3],forward[12:16,:,:,4:7]],axis=-2)
forward_input_depths = tf.concat([tf.expand_dims(forward[12:16,:,:,3],axis=-1),tf.expand_dims(forward[12:16,:,:,7],axis=-1)],axis=-2)
backward_input_images = tf.concat([backward[12:16,:,:,0:3],backward[12:16,:,:,4:7]],axis=-2)
backward_input_depths = tf.concat([tf.expand_dims(backward[12:16,:,:,3],axis=-1),tf.expand_dims(backward[12:16,:,:,7],axis=-1)],axis=-2)
tf.summary.image('input_images_ptb_forward'+summary_type,forward_input_images)
tf.summary.image('input_depths_ptb_forward'+summary_type,forward_input_depths)
tf.summary.image('input_images_ptb_backward'+summary_type,backward_input_images)
tf.summary.image('input_depths_ptb_backward'+summary_type,backward_input_depths)
# middlebury
forward_input_images = tf.concat([forward[16:20,:,:,0:3],forward[16:20,:,:,4:7]],axis=-2)
forward_input_depths = tf.concat([tf.expand_dims(forward[16:20,:,:,3],axis=-1),tf.expand_dims(forward[16:20,:,:,7],axis=-1)],axis=-2)
backward_input_images = tf.concat([backward[16:20,:,:,0:3],backward[16:20,:,:,4:7]],axis=-2)
backward_input_depths = tf.concat([tf.expand_dims(backward[16:20,:,:,3],axis=-1),tf.expand_dims(backward[16:20,:,:,7],axis=-1)],axis=-2)
tf.summary.image('input_images_middlebury_forward'+summary_type,forward_input_images)
tf.summary.image('input_depths_middlebury_forward'+summary_type,forward_input_depths)
tf.summary.image('input_images_middlebury_backward'+summary_type,backward_input_images)
tf.summary.image('input_depths_middlebury_backward'+summary_type,backward_input_depths)
else:
# middlebury
forward_input_images = tf.concat([forward[12:16,:,:,0:3],forward[12:16,:,:,4:7]],axis=-2)
forward_input_depths = tf.concat([tf.expand_dims(forward[12:16,:,:,3],axis=-1),tf.expand_dims(forward[12:16,:,:,7],axis=-1)],axis=-2)
backward_input_images = tf.concat([backward[12:16,:,:,0:3],backward[12:16,:,:,4:7]],axis=-2)
backward_input_depths = tf.concat([tf.expand_dims(backward[12:16,:,:,3],axis=-1),tf.expand_dims(backward[12:16,:,:,7],axis=-1)],axis=-2)
tf.summary.image('input_images_middlebury_forward'+summary_type,forward_input_images)
tf.summary.image('input_depths_middlebury_forward'+summary_type,forward_input_depths)
tf.summary.image('input_images_middlebury_backward'+summary_type,backward_input_images)
tf.summary.image('input_depths_middlebury_backward'+summary_type,backward_input_depths)
def tower_loss(self,scope, images, labels,summary_type='_train'):
"""Calculate the total loss on a single tower running the CIFAR model.
Args:
scope: unique prefix string identifying the CIFAR tower, e.g. 'tower_0'
images: Images. 4D tensor of shape [batch_size, height, width, 3].
labels: Labels. 1D tensor of shape [batch_size].
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
network_input_images, network_input_labels = self.get_network_input_forward(images,labels)
network_input_images_back, network_input_labels_back = self.get_network_input_backward(images,labels)
network_input_images, network_input_labels = self.further_resize_imgs_lbls(network_input_images,network_input_labels)
network_input_images_back, network_input_labels_back = self.further_resize_imgs_lbls(network_input_images_back,network_input_labels_back)
# FB = forward-backward
concatenated_FB_images = tf.concat([network_input_images,network_input_images_back],axis=0)
# Build inference Graph. - forward flow
predict_flows = network.train_network(concatenated_FB_images)
self.write_forward_backward_images(network_input_images,network_input_images_back,network_input_labels,network_input_labels_back,summary_type)
flows_dict = self.get_predict_flow_forward_backward(predict_flows,network_input_labels,concatenated_FB_images,summary_type)
self.write_flows_concatenated_side_by_side(network_input_images,network_input_labels,flows_dict['predict_flow'][0],summary_type)
# # remove middlebury
# network_input_images, network_input_labels = self.remove_mid_records(network_input_images, network_input_labels)
# flows_dict['predict_flow_ref3'][0], flows_dict['predict_flow_ref2'][0] = self.remove_mid_records(flows_dict['predict_flow_ref3'][0], flows_dict['predict_flow_ref2'][0])
# flows_dict['predict_flow_ref1'][0], flows_dict['predict_flow'][0] = self.remove_mid_records(flows_dict['predict_flow_ref1'][0], flows_dict['predict_flow'][0])
# backward_flow_images = losses_helper.forward_backward_loss(predict_flow)
# Build inference Graph. - backward flow
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
# losses sections[self.section_type]
_ = losses_helper.forward_backward_loss(flows_dict['predict_flow_ref3'][0],
flows_dict['predict_flow_ref3'][1],
scope='fb_loss_refine_3'+summary_type)
_ = losses_helper.forward_backward_loss(flows_dict['predict_flow_ref2'][0],
flows_dict['predict_flow_ref2'][1],
scope='fb_loss_refine_2'+summary_type
# ,losses_helper.ease_in_quad(self.global_step,
# 0,
# 1.0,
# 30000.0,
# 50000,
# 'fb_loss_refine_3')
)
_ = losses_helper.forward_backward_loss(flows_dict['predict_flow_ref1'][0],
flows_dict['predict_flow_ref1'][1],
scope='fb_loss_refine_1'+summary_type
# ,losses_helper.ease_in_quad(self.global_step,
# 0,
# 1.0,
# 30000.0,
# 100000,
# 'fb_loss_refine_2')
)
# predict_flow2_label = losses_helper.downsample_label(network_input_labels)
# predict_flow2_refine3 = losses_helper.downsample_label(predict_flows[1],
# size=[20,32],factorU=0.125,factorV=0.125)
# predict_flow2_refine2 = losses_helper.downsample_label(predict_flows[2],
# size=[40,64],factorU=0.25,factorV=0.25)
# predict_flow2_refine1 = losses_helper.downsample_label(predict_flows[3],
# size=[80,128],factorU=0.5,factorV=0.5)
'''
Applying pc loss on lower resolutions
'''
# _ = losses_helper.photoconsistency_loss(network_input_images,flows_dict['predict_flow'][0])
# network_input_images_refine3 = tf.image.resize_images(network_input_images,[20,32],method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)
# network_input_images_refine2 = tf.image.resize_images(network_input_images,[40,64],method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)
# network_input_images_refine1 = tf.image.resize_images(network_input_images,[80,128],method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)
# _ = losses_helper.photoconsistency_loss(network_input_images_refine3,flows_dict['predict_flow_ref3'][0],7,'forward',scope='photoconsistency_loss_refine_forward_3'+summary_type)
# _ = losses_helper.photoconsistency_loss(network_input_images_refine2,flows_dict['predict_flow_ref2'][0],7,'forward',scope='photoconsistency_loss_refine_forward_2'+summary_type)
# _ = losses_helper.photoconsistency_loss(network_input_images_refine1,flows_dict['predict_flow_ref1'][0],7,'forward',scope='photoconsistency_loss_refine_forward_1'+summary_type)
# _ = losses_helper.photoconsistency_loss(network_input_images_refine3,flows_dict['predict_flow_ref3'][1],7,'backward',scope='photoconsistency_loss_refine_backward_3'+summary_type)
# _ = losses_helper.photoconsistency_loss(network_input_images_refine2,flows_dict['predict_flow_ref2'][1],7,'backward',scope='photoconsistency_loss_refine_backward_2'+summary_type)
# _ = losses_helper.photoconsistency_loss(network_input_images_refine1,flows_dict['predict_flow_ref1'][1],7,'backward',scope='photoconsistency_loss_refine_backward_1'+summary_type)
# unsupervised losses done. Now remove ptb. Since it doesn't have ground truth.
if self.section_type == 3:
network_input_images, network_input_labels = self.remove_ptb_records(network_input_images, network_input_labels)
flows_dict['predict_flow_ref3'][0], flows_dict['predict_flow_ref2'][0] = self.remove_ptb_records(flows_dict['predict_flow_ref3'][0], flows_dict['predict_flow_ref2'][0])
flows_dict['predict_flow_ref1'][0], flows_dict['predict_flow'][0] = self.remove_ptb_records(flows_dict['predict_flow_ref1'][0], flows_dict['predict_flow'][0])
# supervised
'''
Applying epe loss on full resolution
'''
_ = losses_helper.endpoint_loss(network_input_labels,flows_dict['predict_flow'][0],scope='epe_loss'+summary_type,summary_type=summary_type)
'''
Applying epe loss on lower resolutions
'''
network_input_labels_refine3 = losses_helper.downsample_label(network_input_labels,
size=[20,32],factorU=0.125,factorV=0.125)
network_input_labels_refine2 = losses_helper.downsample_label(network_input_labels,
size=[40,64],factorU=0.25,factorV=0.25)
network_input_labels_refine1 = losses_helper.downsample_label(network_input_labels,
size=[80,128],factorU=0.5,factorV=0.5)
_ = losses_helper.endpoint_loss(network_input_labels_refine3,flows_dict['predict_flow_ref3'][0],100,scope='epe_loss_refine_3'+summary_type,summary_type=summary_type)
_ = losses_helper.endpoint_loss(network_input_labels_refine2,flows_dict['predict_flow_ref2'][0],100,scope='epe_loss_refine_2'+summary_type,summary_type=summary_type)
_ = losses_helper.endpoint_loss(network_input_labels_refine1,flows_dict['predict_flow_ref1'][0],100,scope='epe_loss_refine_1'+summary_type,summary_type=summary_type)
# _ = losses_helper.photoconsistency_loss(network_input_images,predict_flows[0])
# _ = losses_helper.depth_consistency_loss(network_input_images,predict_flows[0])
# scale_invariant_gradient_image_gt = losses_helper.scale_invariant_gradient(network_input_labels[:,:,:,0:2],
# np.array([1,2,4,8,16]),
# np.array([1,1,1,1,1]))
# scale_invariant_gradient_image_pred = losses_helper.scale_invariant_gradient(flows_dict['predict_flow'][0],
# np.array([1,2,4,8,16]),
# np.array([1,1,1,1,1]))
# _ = losses_helper.scale_invariant_gradient_loss(
# scale_invariant_gradient_image_pred,
# scale_invariant_gradient_image_gt,
# 0.0001,
# self.FLAGS['MAX_STEPS'],
# self.global_step,
# scope='scale_invariant_gradient_loss'+summary_type)
# _ = losses_helper.depth_loss(predict_flow5_label,predict_flow5)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in losses + [total_loss]:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
loss_name = re.sub('%s_[0-9]*/' % 'tower', '', l.op.name)
if summary_type == '_train':
tf.summary.scalar(loss_name+summary_type, l)
elif summary_type =='_test' and not '_train' in loss_name:
tf.summary.scalar(loss_name+summary_type, l)
return total_loss
def get_predict_flow_forward_backward(self,predict_flows,network_input_labels,concatenated_FB_images,summary_type='_train'):
batch_size = predict_flows[0].get_shape().as_list()[0]
batch_half = batch_size // 2
predict_flow = predict_flows[0]
predict_flow_ref1 = predict_flows[1]
predict_flow_ref2 = predict_flows[2]
predict_flow_ref3 = predict_flows[3]
# predict_flow_ref4 = predict_flows[4]
# for other losses, we only consider forward flow
predict_flow_forward = predict_flow[0:batch_half,:,:,:]
predict_flow_backward = predict_flow[batch_half:batch_size,:,:,:]
# predict_flow_forward_ref4 = predict_flow_ref4[0:batch_half,:,:,:]
# predict_flow_backward_ref4 = predict_flow_ref4[batch_half:batch_size,:,:,:]
predict_flow_forward_ref3 = predict_flow_ref3[0:batch_half,:,:,:]
predict_flow_backward_ref3 = predict_flow_ref3[batch_half:batch_size,:,:,:]
predict_flow_forward_ref2 = predict_flow_ref2[0:batch_half,:,:,:]
predict_flow_backward_ref2 = predict_flow_ref2[batch_half:batch_size,:,:,:]
predict_flow_forward_ref1 = predict_flow_ref1[0:batch_half,:,:,:]
predict_flow_backward_ref1 = predict_flow_ref1[batch_half:batch_size,:,:,:]
# tf.summary.image('flow_u_1',network_input_labels[:,:,:,0:1])
# tf.summary.image('flow_v_1',network_input_labels[:,:,:,1:2])
# tf.summary.image('predict_flow_forward_ref4_u'+summary_type,tf.expand_dims(predict_flow_forward_ref4[:,:,:,0],axis=3))
# tf.summary.image('predict_flow_forward_ref4_v'+summary_type,tf.expand_dims(predict_flow_forward_ref4[:,:,:,1],axis=3))
# tf.summary.image('predict_flow_backward_ref4_u'+summary_type,tf.expand_dims(predict_flow_backward_ref4[:,:,:,0],axis=3))
# tf.summary.image('predict_flow_backward_ref4_v'+summary_type,tf.expand_dims(predict_flow_backward_ref4[:,:,:,1],axis=3))
concatenated_fb_ref1_u = tf.concat([tf.expand_dims(predict_flow_forward_ref1[:,:,:,0],axis=-1),tf.expand_dims(predict_flow_backward_ref1[:,:,:,0],axis=-1)],axis=-2)
concatenated_fb_ref1_v = tf.concat([tf.expand_dims(predict_flow_forward_ref1[:,:,:,1],axis=-1),tf.expand_dims(predict_flow_backward_ref1[:,:,:,1],axis=-1)],axis=-2)
# concatenated_fb_ref4_u = tf.concat([tf.expand_dims(predict_flow_forward_ref4[:,:,:,0],axis=-1),tf.expand_dims(predict_flow_backward_ref4[:,:,:,0],axis=-1)],axis=-2)
# concatenated_fb_ref4_v = tf.concat([tf.expand_dims(predict_flow_forward_ref4[:,:,:,1],axis=-1),tf.expand_dims(predict_flow_backward_ref4[:,:,:,1],axis=-1)],axis=-2)
concatenated_fb_ref3_u = tf.concat([tf.expand_dims(predict_flow_forward_ref3[0:4,:,:,0],axis=-1),tf.expand_dims(predict_flow_backward_ref3[0:4,:,:,0],axis=-1)],axis=-2)
concatenated_fb_ref3_v = tf.concat([tf.expand_dims(predict_flow_forward_ref3[0:4,:,:,1],axis=-1),tf.expand_dims(predict_flow_backward_ref3[0:4,:,:,1],axis=-1)],axis=-2)
concatenated_fb_ref2_u = tf.concat([tf.expand_dims(predict_flow_forward_ref2[0:4,:,:,0],axis=-1),tf.expand_dims(predict_flow_backward_ref2[0:4,:,:,0],axis=-1)],axis=-2)
concatenated_fb_ref2_v = tf.concat([tf.expand_dims(predict_flow_forward_ref2[0:4,:,:,1],axis=-1),tf.expand_dims(predict_flow_backward_ref2[0:4,:,:,1],axis=-1)],axis=-2)
concatenated_fb_u = tf.concat([tf.expand_dims(predict_flow_forward[0:4,:,:,0],axis=-1),tf.expand_dims(predict_flow_backward[0:4,:,:,0],axis=-1)],axis=-2)
concatenated_fb_v = tf.concat([tf.expand_dims(predict_flow_forward[0:4,:,:,1],axis=-1),tf.expand_dims(predict_flow_backward[0:4,:,:,1],axis=-1)],axis=-2)
concatenated_fb_u_monkaa = tf.concat([tf.expand_dims(predict_flow_forward[4:8,:,:,0],axis=-1),tf.expand_dims(predict_flow_backward[4:8,:,:,0],axis=-1)],axis=-2)
concatenated_fb_v_monkaa = tf.concat([tf.expand_dims(predict_flow_forward[4:8,:,:,1],axis=-1),tf.expand_dims(predict_flow_backward[4:8,:,:,1],axis=-1)],axis=-2)
concatenated_fb_u_ptb = tf.concat([tf.expand_dims(predict_flow_forward[8:12,:,:,0],axis=-1),tf.expand_dims(predict_flow_backward[8:12,:,:,0],axis=-1)],axis=-2)
concatenated_fb_v_ptb = tf.concat([tf.expand_dims(predict_flow_forward[8:12,:,:,1],axis=-1),tf.expand_dims(predict_flow_backward[8:12,:,:,1],axis=-1)],axis=-2)
# tf.summary.image('concatenated_fb_ref4_u'+summary_type,concatenated_fb_ref4_u)
# tf.summary.image('concatenated_fb_ref4_v'+summary_type,concatenated_fb_ref4_v)
tf.summary.image('concatenated_fb_ref3_u'+summary_type,concatenated_fb_ref3_u)
tf.summary.image('concatenated_fb_ref3_v'+summary_type,concatenated_fb_ref3_v)
tf.summary.image('concatenated_fb_ref2_u'+summary_type,concatenated_fb_ref2_u)
tf.summary.image('concatenated_fb_ref2_v'+summary_type,concatenated_fb_ref2_v)
tf.summary.image('concatenated_fb_final_u'+summary_type,concatenated_fb_u)
tf.summary.image('concatenated_fb_final_v'+summary_type,concatenated_fb_v)
tf.summary.image('concatenated_fb_final_u_monkaa'+summary_type,concatenated_fb_u_monkaa)
tf.summary.image('concatenated_fb_final_v_monkaa'+summary_type,concatenated_fb_v_monkaa)
tf.summary.image('concatenated_fb_final_u_ptb'+summary_type,concatenated_fb_u_ptb)
tf.summary.image('concatenated_fb_final_v_ptb'+summary_type,concatenated_fb_v_ptb)
# tf.summary.image('concatenated_fb_ref1_u'+summary_type,concatenated_fb_ref1_u)
# tf.summary.image('concatenated_fb_ref1_v'+summary_type,concatenated_fb_ref1_v)
# tf.summary.image('predict_flow_forward_ref2_u'+summary_type,tf.expand_dims(predict_flow_forward_ref2[:,:,:,0],axis=3))
# tf.summary.image('predict_flow_forward_ref3_u'+summary_type,tf.expand_dims(predict_flow_forward_ref3[:,:,:,0],axis=3))
# tf.summary.image('predict_flow_forward_ref1_v'+summary_type,tf.expand_dims(predict_flow_forward_ref1[:,:,:,1],axis=3))
# tf.summary.image('predict_flow_forward_ref2_v'+summary_type,tf.expand_dims(predict_flow_forward_ref2[:,:,:,1],axis=3))
# tf.summary.image('predict_flow_forward_ref3_v'+summary_type,tf.expand_dims(predict_flow_forward_ref3[:,:,:,1],axis=3))
# tf.summary.image('predict_flow_backward_ref1_u'+summary_type,tf.expand_dims(predict_flow_backward_ref1[:,:,:,0],axis=3))
# tf.summary.image('predict_flow_backward_ref2_u'+summary_type,tf.expand_dims(predict_flow_backward_ref2[:,:,:,0],axis=3))
# tf.summary.image('predict_flow_backward_ref3_u'+summary_type,tf.expand_dims(predict_flow_backward_ref3[:,:,:,0],axis=3))
# tf.summary.image('predict_flow_backward_ref1_v'+summary_type,tf.expand_dims(predict_flow_backward_ref1[:,:,:,1],axis=3))
# tf.summary.image('predict_flow_backward_ref2_v'+summary_type,tf.expand_dims(predict_flow_backward_ref2[:,:,:,1],axis=3))
# tf.summary.image('predict_flow_backward_ref3_v'+summary_type,tf.expand_dims(predict_flow_backward_ref3[:,:,:,1],axis=3))
# tf.summary.image('input_image1_monkaa'+summary_type,concatenated_FB_images[4:8,:,:,0:3])
# tf.summary.image('input_image2_monkaa'+summary_type,concatenated_FB_images[4:8,:,:,4:7])
# tf.summary.image('depth_image1_monkaa'+summary_type,tf.expand_dims(concatenated_FB_images[4:8,:,:,3],axis=-1))
# tf.summary.image('depth_image2_monkaa'+summary_type,tf.expand_dims(concatenated_FB_images[4:8,:,:,7],axis=-1))
return {
'predict_flow': [predict_flow_forward, predict_flow_backward],
# 'predict_flow_ref4': [predict_flow_forward_ref4,predict_flow_backward_ref4],
'predict_flow_ref3': [predict_flow_forward_ref3,predict_flow_backward_ref3],
'predict_flow_ref2': [predict_flow_forward_ref2, predict_flow_backward_ref2],
'predict_flow_ref1': [predict_flow_forward_ref1, predict_flow_backward_ref1]
}
def write_flows_concatenated_side_by_side(self,network_input_images,network_input_labels,predict_flow2,summary_type='_train'):
concated_flows_u_driving = tf.concat([network_input_labels[0:4,:,:,0:1],predict_flow2[0:4,:,:,0:1]],axis=-2)
concated_flows_v_driving = tf.concat([network_input_labels[0:4,:,:,1:2],predict_flow2[0:4,:,:,1:2]],axis=-2)
denormalized_flow = losses_helper.denormalize_flow(predict_flow2)
warped_img = losses_helper.flow_warp(network_input_images[:,:,:,4:7],denormalized_flow)
tf.summary.image('concated_flows_u_flying'+summary_type,concated_flows_u_driving)
tf.summary.image('concated_flows_v_flying'+summary_type,concated_flows_v_driving)
tf.summary.image('flow_warp_with_original_image_flying'+summary_type,tf.concat([network_input_images[0:4,:,:,0:3],warped_img[0:4,:,:,:]],axis=-2))
if self.section_type > 0 and self.section_type is not 4:
concated_flows_u_flying = tf.concat([network_input_labels[4:8,:,:,0:1],predict_flow2[4:8,:,:,0:1]],axis=-2)
concated_flows_v_flying = tf.concat([network_input_labels[4:8,:,:,1:2],predict_flow2[4:8,:,:,1:2]],axis=-2)
tf.summary.image('concated_flows_u_monkaa'+summary_type,concated_flows_u_flying)