-
Notifications
You must be signed in to change notification settings - Fork 0
/
data_loader.py
executable file
·1455 lines (1151 loc) · 55 KB
/
data_loader.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
"""
data_loader.py
"""
# FROM Python LIBRARY
import os
import random
import numpy as np
import inspect
import itertools
import glob
import time
import multiprocessing
import random
import re
from PIL import Image
from PIL import PngImagePlugin
from PIL import ImageFile
from typing import List, Dict, Tuple, Union, Optional
# FROM PyTorch
import torch
from torch._utils import ExceptionWrapper
from torch.utils.data import Dataset as torchDataset
from torch.utils.data import DataLoader as torchDataLoader
from torch.utils.data._utils import pin_memory
from torch.utils.data.dataloader import _BaseDataLoaderIter, _MultiProcessingDataLoaderIter, _DatasetKind, _SingleProcessDataLoaderIter
from torchvision import transforms
from torchvision import datasets
from torchvision.datasets import ImageFolder
# FROM This Project
import backbone.preprocessing
import backbone.augmentation
import backbone.utils as utils
from backbone.config import Config
# Prevent memory Error in PIL
PngImagePlugin.MAX_TEXT_CHUNK = 100 * (1024 ** 2)
ImageFile.LOAD_TRUNCATED_IMAGES = True
# Read Preprocessings from backbone.preprocessing and Augmentations from backbone.augmentation automatically
PREPROCESSING_DICT = dict(
x for x in inspect.getmembers(backbone.preprocessing) if (not x[0].startswith("__")) and (inspect.isclass(x[1]))
)
AUGMENTATION_DICT = dict(
x for x in inspect.getmembers(backbone.augmentation) if (not x[0].startswith("_")) and (inspect.isfunction(x[1]))
)
class DatasetConfig:
def __init__(
self,
name: str,
origin: Optional[str] = None,
dataType: Optional[Dict[str, str]] = None,
labelType: Optional[Dict[str, str]] = None,
availableMode: Optional[List[str]] = None,
classes: Optional[List[str]] = None,
preprocessings: Optional[List[str]] = None,
useDatasetConfig: bool = True,
):
self.name = name
self.origin = origin
self.dataType = dataType
self.labelType = labelType
self.availableMode = availableMode
self.classes = classes
self.preprocessings = preprocessings
self.useDatasetConfig = useDatasetConfig
if self.useDatasetConfig is True:
self._getDatasetConfig()
def _splitDataType(self, dataType: Dict[str, str]):
# return form of {'dataName': 'LR', dataType': 'text', 'tensorType': 'int'}
dataTypeList = ["Text", "Image", "ImageSequence"]
tensorTypeList = ["Float", "Int", "Long", "Double"]
if dataType is not None:
for key in dataType:
assert (
dataType[key].split("-")[0] in dataTypeList
), f"data_loader.py :: dataset config {self.name} has invalid dataType."
assert (
dataType[key].split("-")[1] in tensorTypeList
), f"data_loader.py :: dataset config {self.name} has invalid tensorType."
return dict(zip(["dataName", "dataType", "tensorType"], [key] + dataType[key].split("-")))
else:
return {}
def _getDatasetConfig(self):
yamlData = Config.datasetConfigDict[f"{self.name}"]
self.origin = str(yamlData["origin"])
self.dataType = self._splitDataType(yamlData["dataType"])
self.labelType = self._splitDataType(yamlData["labelType"])
self.availableMode = list(map(str, yamlData["availableMode"]))
self.classes = list(map(str, yamlData["classes"]))
self.preprocessings = list(map(str, yamlData["preprocessings"])) if yamlData["preprocessings"] is not None else []
# Representation of single dataset property
class DatasetComponent:
def __init__(self, name, mode, classParameter, labelClassName, sequenceLength):
self.name = name
self.datasetConfig = None
self._getDatasetConfigByComponentName()
self.mode = mode
self.classParameter = classParameter
self.labelClassName = labelClassName
self.sequenceLength = sequenceLength
self.dataFileList = None
self.labelFileList = None
self._getDataFileList()
self.preprocessingList = None
self._makePreprocessingList()
def _getDatasetConfigByComponentName(self):
self.datasetConfig = DatasetConfig(Config.paramDict["data"]["datasetComponent"][self.name]["dataConfig"])
def _getDataFileList(self):
# get origin path
mainPath = Config.param.data.path.datasetPath
path = f"{self.datasetConfig.origin}/{self.mode}/"
datasetConfigClasses = self.datasetConfig.classes
# dynamic construction of class path based on defined classes
for i in range(len(datasetConfigClasses)):
classPathList = (
list(
itertools.chain.from_iterable(
list(
map(
lambda y: list(
map(
lambda x: str(x) if type(x) is int else x + "/" + str(y) if type(y) is int else y,
classPathList,
)
),
self.classParameter[datasetConfigClasses[i]],
)
)
)
)
if i is not 0
else self.classParameter[datasetConfigClasses[i]]
)
# Sorry for who read this
# add origin path in front of all elements of class path list
pathList = list(map(lambda x: path + x, classPathList))
if self.datasetConfig.dataType["dataType"] == "Text":
# construct all of readable file lists in class path lists
dataFileLists = list(
map(
lambda x: list(
filter(
lambda x: (x.endswith(".txt")),
list(map(lambda y: mainPath + x + "/" + y, sorted(os.listdir(mainPath + x)))),
)
),
pathList,
)
)
elif self.datasetConfig.dataType["dataType"] == "Image":
# construct all of readable file lists in class path lists
dataFileLists = list(
map(
lambda x: list(
filter(
lambda x: (
x.endswith(".png") or x.endswith(".jpg") or x.endswith(".jpeg") or x.endswith(".bmp")
),
list(map(lambda y: mainPath + x + "/" + y, sorted(os.listdir(mainPath + x)))),
)
),
pathList,
)
)
elif self.datasetConfig.dataType["dataType"] == "ImageSequence":
# construct all of sequence folders in class path lists
dataFileLists = list(map(lambda x: list(map(lambda y: x + "/" + y, sorted(os.listdir(mainPath + x)))), pathList))
dataFileLists = [
list(
map(
lambda x: [
mainPath + x + "/" + z
for z in list(
filter(
lambda y: (
y.endswith(".png") or y.endswith(".jpg") or y.endswith(".jpeg") or y.endswith(".bmp")
),
sorted(os.listdir(mainPath + x)),
)
)
],
xlist,
)
)
for xlist in dataFileLists
]
# if label Exists
if len(self.datasetConfig.labelType) > 0:
labelPath = f"{path}{self.labelClassName}/"
if self.datasetConfig.labelType["dataType"] == "Text":
# construct all of readable file lists in class path lists
labelFiles = sorted(list(filter(lambda x: (x.endswith(".txt")), os.listdir(mainPath + labelPath))))
# add origin path in front of all elements of label file list
labelFiles = list(map(lambda x: mainPath + labelPath + x, labelFiles))
elif self.datasetConfig.labelType["dataType"] == "Image":
# construct all of readable file lists in class path lists
labelFiles = sorted(
list(
filter(
lambda x: (
x.endswith(".png") or x.endswith(".jpg") or x.endswith(".jpeg") or x.endswith(".bmp")
),
os.listdir(mainPath + labelPath),
)
)
)
# add origin path in front of all elements of label file list
labelFiles = list(map(lambda x: mainPath + labelPath + x, labelFiles))
elif self.datasetConfig.labelType["dataType"] == "ImageSequence":
# construct all of sequence folders in class path lists
labelFiles = sorted(os.listdir(mainPath + labelPath))
labelFiles = list(
map(
lambda x: [
mainPath + labelPath + x + "/" + z
for z in list(
filter(
lambda y: (
y.endswith(".png") or y.endswith(".jpg") or y.endswith(".jpeg") or y.endswith(".bmp")
),
sorted(os.listdir(mainPath + labelPath + x)),
)
)
],
labelFiles,
)
)
# print(labelFiles)
if len(labelFiles) == 1:
# Label case ex (.txt file)
labelFiles = labelFiles * len(dataFileLists[0])
else:
# Label case for same count of image files (GT)
assert [len(dataFileLists[0])] * len(dataFileLists) == list(
map(len, dataFileLists)
), f"data_loader.py :: ERROR! dataset {self.name} has NOT same count of data files for each classes."
assert [len(labelFiles)] * len(dataFileLists) == list(
map(len, dataFileLists)
), f"data_loader.py :: ERROR! label and data files should be had same count. (dataset {self.name})"
else:
labelFiles = [None] * len(dataFileLists[0])
labelFiles = [labelFiles] * len(dataFileLists)
dataFileDictList = list(
map(
lambda x: dict(zip(["dataFilePath", "labelFilePath"], x)),
list(zip(itertools.chain.from_iterable(dataFileLists), itertools.chain.from_iterable(labelFiles))),
)
)
# set dataFileList without main path
self.dataFileList = dataFileDictList
def _makePreprocessingList(self):
self.preprocessingList = list(
map(
(
lambda x: PREPROCESSING_DICT[x.split("(")[0]](
*list(filter(lambda y: y != "", x.split("(")[1][:-1].replace(" ", "").split(",")))
)
),
self.datasetConfig.preprocessings,
)
)
def _getSeqDataLen(self):
cnt = 0
for seqFileDict in self.dataFileList:
# for key in seqFileDict:
cnt += len(seqFileDict[list(seqFileDict.keys())[0]]) - (self.sequenceLength - 1)
# print(cnt)
return cnt
def __len__(self):
return (
len(self.dataFileList) if self.datasetConfig.dataType["dataType"] != "ImageSequence" else self._getSeqDataLen()
)
# TODO: Distribution
class Dataset(torchDataset):
def __init__(
self,
datasetComponentParamList: List,
batchSize: int,
valueRangeType: str,
isEval: bool,
augmentation: List[str],
numWorkers: int,
makePreprocessedFile: bool,
isCaching: bool,
forceImageDataType: bool,
):
'''
Dataset Settings
'''
self.datasetComponentParamList = datasetComponentParamList
self.datasetComponentObjectList = None
self._constructDatasetComponents()
self._datasetComponentObjectListIntegrityTest()
self.batchSize = batchSize
self.valueRangeType = valueRangeType
self.isEval = isEval
self.augmentation = augmentation
self.numWorkers = numWorkers
self.makePreprocessedFile = makePreprocessedFile
self.PILImageToTensorFunction = transforms.ToTensor()
self.isCaching = isCaching
self.forceImageDataType = forceImageDataType
'''
Dataset data
'''
self.globalFileList = None
self._makeGlobalFileList()
self.mapper = None
self._makeFileListIndexer()
self.LabelDataDictList = None
self.cache = None#{}#FILE_CACHE
if self.isCaching is True:
self._caching()
def _cachingFunc(self, zippedLists, isMultiProc = True):
if isMultiProc is True:
cache = {}
for filePath, preProc in zippedLists:
cache.update({filePath: self._readImage(filePath, preProc)})
self.cache_cnt.value += 1
print(f"ETA : {(self.cache_MX_LEN - self.cache_cnt.value) / self.cache_cnt.value * (time.perf_counter() - self.cache_startTime):.0f} sec [{self.cache_cnt.value} / {self.cache_MX_LEN}] ", end='\r')
else:
cache = {}
cnt = 0
for filePath, preProc in zippedLists:
cache.update({filePath: self._readImage(filePath, preProc)})
cnt += 1
print(f"ETA : {(self.cache_MX_LEN - cnt) / cnt * (time.perf_counter() - self.cache_startTime):.0f} sec [{cnt} / {self.cache_MX_LEN}] ", end='\r')
return cache
def _caching(self):
datasetComponentType = self.datasetComponentObjectList[0].datasetConfig.dataType["dataType"]
t = time.perf_counter()
MULTIPROC = 0
print("Make cache data list...")
fileListDict = {}
cnt = 0
for dCO in self.datasetComponentObjectList:
print(f"Caching Dataset '{dCO.datasetConfig.name}' ... ")
preProc = dCO.preprocessingList
for i, dfDict in enumerate(dCO.dataFileList):
for dfDictKey in dfDict.keys():
assert datasetComponentType in ['Image', 'ImageSequence']
if datasetComponentType == 'Image':
filePath = dfDict[dfDictKey]
fileListDict.update({filePath:preProc})
elif datasetComponentType == 'ImageSequence':
for filePath in dfDict[dfDictKey]:
fileListDict.update({filePath:preProc})
cnt += 1
print(f"Make list... {cnt} ", end='\r')
fileList = list( fileListDict.keys() )
preProcList = list( fileListDict[x] for x in fileList )
print(f"Start Caching with {MULTIPROC} processes...")
self.cache = {}
MX_LEN = len(fileList)
self.cache_MX_LEN = MX_LEN
self.cache_startTime = time.perf_counter()
if MULTIPROC == 0:
self.cache = self._cachingFunc(zip(fileList,preProcList), isMultiProc=False)
else:
CHNK_SIZE = MX_LEN // MULTIPROC
cacheeList = []
for i in range(MULTIPROC):
stIdx = i * CHNK_SIZE
edIdx = (i + 1) * CHNK_SIZE if i != MULTIPROC - 1 else MX_LEN
cacheeList.append(list(zip(fileList[stIdx: edIdx],preProcList[stIdx: edIdx])))
with multiprocessing.Manager() as manager:
self.cache_cnt = manager.Value(int, 0)
pool = multiprocessing.Pool(processes=MULTIPROC + 1)
[self.cache.update(x) for x in pool.map(self._cachingFunc, cacheeList)]
pool.close()
pool.join()
print(f"Cached in {time.perf_counter() - t:.1f} sec. ")
def _makeGlobalFileList(self):
gFL = [x.dataFileList for x in self.datasetComponentObjectList]
self.globalFileList = gFL
def _makeLabelDataDict(self, labelPath):
"""
Arguments:
labelPath: (string) text file path of images and annotations
Return:
A dict containing:
file path, label value
1) string
2) list of float
intput : label path
output : dict (file path(str), label value[list of list])
"""
labels = []
LabelDict = {}
labels_copy = []
fullPath = ""
isFirst = True
upperName = labelPath.split("/")[-2] # GT
modeName = labelPath.split("/")[-1] # label_train
f = open(labelPath, "r")
lines = f.readlines()
for line in lines:
line = line.rstrip()
if line.startswith("#"):
if isFirst is True:
isFirst = False
else:
labels_copy = labels.copy()
LabelDict.update({fullPath: labels_copy})
labels.clear()
path = line[2:]
fullPath = labelPath.replace(upperName + "/" + modeName, "") + path
else:
line = line.split(" ")
label = [float(x) for x in line]
labels.append(label)
labels_copy = labels.copy()
LabelDict.update({fullPath: labels_copy})
return LabelDict
def _makeGlobalLabelDataDictList(self):
"""
# Commponent Count 후 각 Commponent에 접근의 Label에 접근
# output : list [dict, dict, ..., dict] # dict{filepath : values}
# firstIndex -> Component에 대한 type는 동일하게만 들어옴 (txt)
"""
if self.datasetComponentObjectList[0].datasetConfig.dataType["dataType"] == "Text":
firstIndex = 0
LabelDataPathList = []
componentList = len(self.datasetComponentObjectList)
LabelDataPathList = list(
map(
lambda x: Config.param.data.path.datasetPath + self.globalFileList[x][firstIndex]["labelFilePath"],
range(componentList),
)
)
LabelDataDictList = list(map(lambda x: self._makeLabelDataDict(x), LabelDataPathList))
self.LabelDataDictList = LabelDataDictList
else:
self.LabelDataDictList = {}
def _makeFileListIndexer(self):
mapper = []
dcCount = 0 # datasetComponent의 갯수
dcLenFilesList = [] # datasetComponent 각각이 가지고 있는 총 파일의 갯 수 리스트
dcLenSeqsList = [] # datasetComponent가 가지고 있는 각각의 시퀀스 갯수 리스트
SeqLenFilesList = [] # 각 dc의 시퀀스 밑에 몇개의 파일이 있는지 리스트 (2-d)
####################################################################
# SEQUENCE
####################################################################
if self.datasetComponentObjectList[0].datasetConfig.dataType["dataType"] == "ImageSequence":
for dcL in self.globalFileList:
dcCount += 1
dcLenSeqsList.append(len(dcL))
tempLenFiles = 0
tempSeqLenFiles = []
for seqL in dcL:
ln = len(seqL[list(seqL.keys())[0]]) - (self.datasetComponentObjectList[0].sequenceLength - 1)
tempSeqLenFiles.append(ln)
tempLenFiles += ln
SeqLenFilesList.append(tempSeqLenFiles)
dcLenFilesList.append(tempLenFiles)
####################################################################
# EVAL MODE : len -> sum(len(datasetComponents))
####################################################################
if self.isEval is True:
######## SEQ [ [dc [seq [file, file] ],[seq] ],[dc] ]
for i in range(self.__len__()):
for j in range(len(dcLenFilesList)):
if i < sum(dcLenFilesList[: j + 1]):
break
dcIdx = j
tMapperElem = []
tmp = i // dcCount % dcLenFilesList[dcIdx]
for j in range(len(SeqLenFilesList[dcIdx]) - 1):
if tmp < sum(SeqLenFilesList[dcIdx][: j + 1]):
break
seqIdx = j
fileIdx = tmp - sum(SeqLenFilesList[dcIdx][:j])
for s in range(self.datasetComponentObjectList[dcIdx].sequenceLength):
tMapperElem.append([dcIdx, seqIdx, fileIdx + s])
mapper.append(tMapperElem)
####################################################################
# TRAIN MODE : len -> max(len(datasetComponents)) * 2
####################################################################
else:
for i in range(self.__len__()):
tMapperElem = []
dcIdx = i % dcCount
tmp = i // dcCount % dcLenFilesList[dcIdx]
for j in range(len(SeqLenFilesList[dcIdx])):
if tmp < sum(SeqLenFilesList[dcIdx][: j + 1]):
break
seqIdx = j
fileIdx = tmp - sum(SeqLenFilesList[dcIdx][:j])
for s in range(self.datasetComponentObjectList[dcIdx].sequenceLength):
tMapperElem.append([dcIdx, seqIdx, fileIdx + s])
mapper.append(tMapperElem)
####################################################################
# Non-SEQUENCE
####################################################################
else:
for dcL in self.globalFileList:
dcCount += 1
dcLenFilesList.append(len(dcL))
####################################################################
# EVAL MODE : len -> sum(len(datasetComponents))
####################################################################
if self.isEval is True:
for i in range(self.__len__()):
for j in range(len(dcLenFilesList)):
if i < sum(dcLenFilesList[: j + 1]):
break
dcIdx = j
fileIdx = i - sum(dcLenFilesList[:j])
mapper.append([dcIdx, fileIdx])
####################################################################
# TRAIN MODE : len -> max(len(datasetComponents)) * 2
####################################################################
else:
for i in range(self.__len__()):
dcIdx = i % dcCount
fileIdx = i // dcCount % dcLenFilesList[dcIdx]
mapper.append([dcIdx, fileIdx])
self.mapper = mapper
# print(self.mapper)
# print("str(len(self.mapper))" + str(len(self.mapper)))
def _popItemInGlobalFileListByIndex(self, index):
datasetComponentType = self.datasetComponentObjectList[0].datasetConfig.dataType["dataType"]
if datasetComponentType != "ImageSequence":
componentIndex, componentFileListIndex = self.mapper[index]
# print(componentIndex, componentFileListIndex)
# componentFileListIndex = ( index // len(datasetComponentLengthList) ) % datasetComponentLengthList[componentIndex]
dFP = self.globalFileList[componentIndex][componentFileListIndex]["dataFilePath"]
dFP = (
dFP
if isinstance(dFP, str)
else [x for x in dFP]
)
lFP = (
self.globalFileList[componentIndex][componentFileListIndex]["labelFilePath"]
if self.globalFileList[componentIndex][componentFileListIndex]["labelFilePath"] is not None
else None
)
lFP = (
(
lFP
if isinstance(lFP, str)
else [x for x in lFP]
)
if lFP is not None
else None
)
else:
dFPList = []
lFPList = []
for componentIndex, seqIdx, fileIdx in self.mapper[index]:
"""
datasetComponentSequenceLengthList = []
for dComponent in range(len(self.globalFileList)):
tdC = []
for dSeq in range(len(self.globalFileList[dComponent])):
tdC.append(len(self.globalFileList[dComponent][dSeq]))
datasetComponentSequenceLengthList.append(tdC)
sequenceListIndex = ( index // len(datasetComponentLengthList) ) % datasetComponentLengthList[componentIndex]
sequenceFileIndex =
"""
# print(componentIndex, seqIdx, fileIdx)
dFP = self.globalFileList[componentIndex][seqIdx]["dataFilePath"][fileIdx]
dFP = (
dFP
if isinstance(dFP, str)
else [x for x in dFP]
)
lFP = (
self.globalFileList[componentIndex][seqIdx]["labelFilePath"][fileIdx]
if self.globalFileList[componentIndex][seqIdx]["labelFilePath"] is not None
else None
)
lFP = (
(
lFP
if isinstance(lFP, str)
else [x for x in lFP]
)
if lFP is not None
else None
)
dFPList.append(dFP)
lFPList.append(lFP)
dFP = dFPList
lFP = lFPList
return {"dataFilePath": dFP, "labelFilePath": lFP}
def _constructDatasetComponents(self):
self.datasetComponentObjectList = [DatasetComponent(*x) for x in self.datasetComponentParamList]
def _datasetComponentObjectListIntegrityTest(self):
# Test All dataComponents have same
# dataType
# TensorType
# name
datasetComponentdataTypeList = [x.datasetConfig.dataType for x in self.datasetComponentObjectList]
assert [datasetComponentdataTypeList[0]] * len(
datasetComponentdataTypeList
) == datasetComponentdataTypeList, "data_loader.py :: All datasets in dataloader must have same dataType."
datasetComponentlabelTypeList = [x.datasetConfig.labelType for x in self.datasetComponentObjectList]
assert [datasetComponentlabelTypeList[0]] * len(
datasetComponentlabelTypeList
) == datasetComponentlabelTypeList, "data_loader.py :: All datasets in dataloader must have same tensorType."
def _calculateMemorySizePerTensor(
self, dtype: torch.dtype, expectedShape: List[int]
): # WARN: EXCEPT BATCH SIIZEEEEEE!!!!!!!!!!!!!!!!!!!!!!!
sizeOfOneElementDict = {
torch.float32: 4,
torch.float: 4,
torch.float64: 8,
torch.double: 8,
torch.float16: 2,
torch.half: 2,
torch.uint8: 1,
torch.int8: 1,
torch.int16: 2,
torch.short: 2,
torch.int32: 4,
torch.int: 4,
torch.int64: 8,
torch.long: 8,
torch.bool: 2,
}
elemSize = sizeOfOneElementDict(dtype)
totalSize = reduce(lambda x, y: x * y, expectedShape) * elemSize
return totalSize
def _torchvisionPreprocessing(self, pilImage, preProc):
x = pilImage
for preProcFunc in preProc:
x = preProcFunc(x)
return x
def _applyAugmentationFunction(self, tnsr, augmentationFuncStr: str):
return _applyAugmentationFunctionFunc(tnsr, augmentationFuncStr)
# DATA AUGMENTATION ON CPU - Multiprocessing with Forked Worker processes -
# Before toTensor() Augmentation
# CROP OPERATION usually be here
# Make Sure that ALL output Size (C, H, W) are same.
def _dataAugmentation(self, tnsrList, augmentations: List[str]):
# If input is a list of two tensors -> make it to list of list of two tensors
# The standard Input type is list of list of two tensors -> [ [data_1, label_1], ... , [data_N, label_N] ]
if not isinstance(tnsrList[0], list):
xList = [tnsrList]
else:
xList = tnsrList
augedXList = []
for x in xList:
for augmentation in augmentations:
#print(f"RUN PHASE 3-4-1-{augmentation}")
x = self._applyAugmentationFunction(x, augmentation)
# print(augmentation)
if augmentation == "toTensor()":
break
augedXList.append(x)
# TRANSPOSE
augedXList = list(map(list, zip(*augedXList)))
augedXTensorList = [torch.stack(x) for x in augedXList]
return augedXTensorList
def _setTensorValueRange(self, tnsr, valueRangeType: str):
if valueRangeType == '-1~1':
tnsr = tnsr * 2 - 1
elif valueRangeType == '0~255':
tnsr = np.round(tnsr * 255).type(torch.FloatTensor)
return tnsr
def _loadPILImagesFromHDD(self, filePath):
return Image.open(filePath)
def _saveTensorsToHDD(self, tnsr, filePath):
print(f"Write Tensor to {filePath}.npy...")
utils.saveTensorToNPY(tnsr, filePath)
def _loadNPArrayFromHDD(self, filePath):
return utils.loadNPY(filePath)
def _PIL2Tensor(self, pilImage):
return self.PILImageToTensorFunction(pilImage)
def _NPYMaker(self, filePath, preProc):
PILImage = self._loadPILImagesFromHDD(filePath) # PIL
PPedPILImage = self._torchvisionPreprocessing(PILImage, preProc)
rstTensor = self._PIL2Tensor(PPedPILImage)
self._saveTensorsToHDD(rstTensor, filePath)
return rstTensor
def _methodNPYExists(self, filePath):
npa = self._loadNPArrayFromHDD(filePath)
return npa
def _methodNPYNotExists(self, filePath, preProc):
PILImage = self._loadPILImagesFromHDD(filePath)
PPedPILImage = self._torchvisionPreprocessing(PILImage, preProc)
return PPedPILImage
def _methodLoadLabel(self, componentIndex, filePath, preProc):
labelValueList = self.LabelDataDictList[componentIndex][filePath] # label values(list(list))
for preProcFunc in preProc:
LabelList = preProcFunc(labelValueList) # preProc(labelValueList)
return LabelList
def _readImage(self, FilePath, preProc):
'''
if os.path.isfile(FilePath + ".npy") is True:
try:
rst = self._methodNPYExists(
FilePath + ".npy"
) # if .npy Exists, load preprocessed .npy File as Pytorch Tensor -> load to GPU directly -> Augmentation on GPU -> return
except ValueError:
if self.makePreprocessedFile is True:
rst = self._NPYMaker(
FilePath, preProc
) # if .npy doesn't Exists and self.makePreprocessedFile is True, make .npy file and augmentating tensor and return
else:
rst = self._methodNPYNotExists(
FilePath, preProc
) # if .npy doesn't Exists, load Image File as PIL Image -> Preprocess PIL Image on CPU -> convert to Tensor -> load to GPU -> Augmentation on GPU -> return
else:
if self.makePreprocessedFile is True:
rst = self._NPYMaker(
FilePath, preProc
) # if .npy doesn't Exists and self.makePreprocessedFile is True, make .npy file and augmentating tensor and return
else:
rst = self._methodNPYNotExists(
FilePath, preProc
) # if .npy doesn't Exists, load Image File as PIL Image -> Preprocess PIL Image on CPU -> convert to Tensor -> load to GPU -> Augmentation on GPU -> return
'''
rst = self._methodNPYNotExists(FilePath, preProc)
return rst
def _readItem(self, Type, FilePath, index, preProc):
# data information defined in config
if FilePath is not None:
###################################################################################
# CASE TEXT
###################################################################################
if Type["dataType"] == "Text":
componentIndex = index % len(self.datasetComponentObjectList)
popped = self._popItemInGlobalFileListByIndex(index)
dataFilePath = popped["dataFilePath"]
rst = self._methodLoadLabel(componentIndex, dataFilePath, preProc)
###################################################################################
# CASE IMAGE
###################################################################################
elif Type["dataType"] == "Image":
rst = self._readImage(FilePath, preProc) if self.isCaching is False else self.cache[FilePath]
###################################################################################
# CASE IMAGESEQUENCE
###################################################################################
elif Type["dataType"] == "ImageSequence":
seqFileList = FilePath
rstList = []
for seqFilePath in seqFileList:
rstList.append(self._readImage(seqFilePath, preProc) if self.isCaching is False else self.cache[seqFilePath])
rst = rstList
# data information not defined in config
else:
rst = None
return rst
def __getitem__(self, index):
#print(f"RUN PHASE 3-1")
# popping File Path at GFL(self.globalFileList) by index
popped = self._popItemInGlobalFileListByIndex(index)
dataFilePath = popped["dataFilePath"]
labelFilePath = popped["labelFilePath"]
#print(f"RUN PHASE 3-2")
componentIndex = index % len(self.datasetComponentObjectList)
preProc = self.datasetComponentObjectList[componentIndex].preprocessingList
dataType = self.datasetComponentObjectList[componentIndex].datasetConfig.dataType
labelType = self.datasetComponentObjectList[componentIndex].datasetConfig.labelType
rstDict = {}
filePathList = [dataFilePath] if labelFilePath is None else [dataFilePath, labelFilePath]
typeList = [dataType] if labelType == {} else [dataType, labelType]
#print(f"RUN PHASE 3-3")
###################################################################################
# ADD DATA & LABEL
###################################################################################
for Type, FilePath in zip(typeList, filePathList):
rstDict[Type["dataName"]] = self._readItem(Type, FilePath, index, preProc)
#print(f"RUN PHASE 3-4")
###################################################################################
# Data Augmentation
###################################################################################--------------------------------------------------------------
if isinstance(rstDict[dataType["dataName"]], list): # TODO:
if labelType != {}: # if label Exists
rstDict[dataType["dataName"]], rstDict[labelType["dataName"]] = list(
zip(
*list(
self._dataAugmentation(x, self.augmentation)
for x in list(zip(rstDict[dataType["dataName"]], rstDict[labelType["dataName"]]))
)
)
)
rstDict[labelType["dataName"]] = list(
self._setTensorValueRange(x, self.valueRangeType) for x in rstDict[labelType["dataName"]]
)
else: # if label not Exists
rstDict[dataType["dataName"]] = list(
self._dataAugmentation(x, self.augmentation)[0] for x in list(zip(rstDict[dataType["dataName"]]))
)
rstDict[dataType["dataName"]] = list(
self._setTensorValueRange(x, self.valueRangeType) for x in rstDict[dataType["dataName"]]
)
else:
if labelType != {}: # if label Exists
#print(f"RUN PHASE 3-4-1")
rstDict[dataType["dataName"]], rstDict[labelType["dataName"]] = self._dataAugmentation(
[rstDict[dataType["dataName"]], rstDict[labelType["dataName"]]], self.augmentation