/
data_generator.py
143 lines (114 loc) · 5.12 KB
/
data_generator.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
import logging
import numpy as np
import cv2
import uuid
from keras.preprocessing.image import ImageDataGenerator
import tools
from labels import Labels
class DataGenerator:
MODES = tools.MODES
GROUPS = tools.GROUPS
def __init__(self, config, image_dir=None, labels_json=None, mode=None, group=None, class_counts=None):
self.log = logging.getLogger('root')
self.config = config
self.image_dir = tools.str2path(image_dir)
self.labels_json = tools.str2path(labels_json)
self.mode = mode
self.group = group
self.channels = config.get('channels', 3)
self._batch_size = config.get('flow', {}).get('batch_size', 4)
self._validate_input()
self.color_mode = 'rgb' if self.channels == 3 else 'grayscale'
self.target_imsize = (self.config['target_height'], self.config['target_width'])
self.blacklist = self.config.get('class_blacklist', [])
self.class_counts = class_counts
if labels_json:
self.keras_dir = self.image_dir.parent / f'{self.image_dir.name}_{self.group}_{str(uuid.uuid1())[:8]}'
if not self.keras_dir.exists():
self.keras_dir.mkdir()
self.log.info(f'Creating Keras Directory Tree - {self.mode}...')
tools.create_keras_image_directory_tree(self.image_dir,
self.keras_dir,
self.labels_json,
self.group,
self.blacklist,
self.class_counts)
else:
self.log.info('Skipped Keras Directory Tree creation as it already exists')
self.labels = Labels(labels_json, self.keras_dir, group)
else:
self.keras_dir = image_dir
self._data_generator = self._init_data_gen()
self._flow_gen = self._flow_from_directory()
@property
def data_generator(self):
return self._data_generator
@property
def samples(self):
return len(list(self.keras_dir.rglob("*.jpg")))
@property
def batch_size(self):
return self._batch_size
@property
def steps_per_epoch(self):
return np.floor(self.samples / self.batch_size)
@property
def flow_generator(self):
return self._flow_gen
@property
def class_weights(self):
if self.config.get('weights_type', 'frequency') == 'frequency':
scales = self.labels.inverse_frequency
else:
scales = self.labels.proportions
cls2idx = self._flow_gen.class_indices
class_weights = dict()
for cls, idx in cls2idx.items():
class_weights[idx] = scales[cls]
return class_weights
def _validate_input(self):
if not self.image_dir.exists():
raise IOError(f'{self.image_dir} does not exist')
if self.labels_json is not None and not self.labels_json.exists():
raise IOError(f'{self.labels_json} does not exist')
if self.group is not None and self.group not in self.GROUPS:
raise AttributeError(f'{self.group} is unknown group, known are {self.GROUPS}')
if self.channels != 3 and self.channels != 1:
raise AttributeError(f'channels must be either 3 or 1')
def _init_data_gen(self):
cfg = self.config.get(f'{self.mode}_gen')
if cfg:
return ImageDataGenerator(**cfg)
else:
return ImageDataGenerator()
def _flow_from_directory(self):
cfg = self.config.get(f'flow', {})
return self._data_generator.flow_from_directory(directory=self.keras_dir,
target_size=self.target_imsize,
color_mode=self.color_mode,
**cfg)
def flow_from_labels(self):
ims_in_dir = list(self.keras_dir.rglob("*.jpg"))
ims_in_dir_names = [file.name for file in ims_in_dir]
images = list()
labels = list()
image_names = list()
while len(ims_in_dir_names):
image_name = ims_in_dir_names.pop(0)
image_path = ims_in_dir.pop(0)
idx = np.where(self.labels.image_names==image_name)[0]
images.append(self._load_image(image_path))
labels.append(self.labels.numerical[idx])
image_names.append(image_name)
if len(image_names) == self.batch_size:
labels = np.array(labels)
yield np.array(images), np.reshape(labels, (len(labels))), image_names
images = list()
labels = list()
image_names = list()
if len(image_names):
labels = np.array(labels)
yield np.array(images), np.reshape(labels, (len(labels))), image_names
def _load_image(self, path):
grayscale = self.channels == 1
return tools.load_image(path, (self.target_imsize[1], self.target_imsize[0]), grayscale)