-
Notifications
You must be signed in to change notification settings - Fork 0
/
data.py
187 lines (140 loc) · 6.92 KB
/
data.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
import os
import cv2
import numpy as np
import tensorflow as tf
from natsort import natsorted
from sklearn.utils import shuffle
from matplotlib import pyplot as plt
from albumentations import (Compose, GaussianBlur, MedianBlur, CLAHE, Emboss,
RandomBrightnessContrast, RandomCrop, RandomRotate90,
GaussNoise, HorizontalFlip, VerticalFlip)
def transform():
"""
Create and return a list of image augmentation operations to be applied during the training process.
Returns:
list: A list of image augmentation operations to be applied on images.
"""
return Compose([
GaussianBlur(always_apply=False, p=0.25, blur_limit=(3, 7), sigma_limit=(0.0, 0)),
MedianBlur(always_apply=False, p=0.25, blur_limit=(3, 7)),
CLAHE(always_apply=False, p=0.25, clip_limit=(1, 3), tile_grid_size=(4, 4)),
Emboss(always_apply=False, p=0.25, alpha=(0.39, 0.45), strength=(0.2, 2.04)),
RandomBrightnessContrast(always_apply=False, p=0.25, brightness_limit=(-0.2, 0.2), contrast_limit=(-0.2, 0.2), brightness_by_max=True),
RandomCrop(always_apply=False, p=0.25, height=220, width=220),
RandomRotate90(always_apply=False, p=0.25),
GaussNoise(always_apply=False, p=0.25, var_limit=(92.11, 149.67), per_channel=True, mean=0.0),
HorizontalFlip(always_apply=False, p=0.25),
VerticalFlip(always_apply=False, p=0.25),
])
class DataGenerator(tf.keras.utils.Sequence):
"""
Custom data generator for training process using image and mask data.
Args:
root_dir (str): Root directory path.
image_folder (str): Folder name containing the images.
mask_folder (str): Folder name containing the masks.
image_size (int, optional): Size of the images. Defaults to 256.
batch_size (int, optional): Batch size. Defaults to 4.
transform (callable, optional): Image transformation function. Defaults to None.
shuffle (bool, optional): Whether to shuffle the data. Defaults to True.
"""
def __init__(self,
root_dir,
image_folder,
mask_folder,
image_size=256,
batch_size=4,
transform=None,
shuffle=True):
super(DataGenerator, self).__init__()
self.path = root_dir
self.image_names = natsorted(next(os.walk(os.path.join(root_dir, image_folder)))[2])
self.mask_names = natsorted(next(os.walk(os.path.join(root_dir, mask_folder)))[2])
self.image_size = image_size
self.batch_size = batch_size
self.currentIndex = 0
self.indexes = None
self.transform = transform
self.shuffle = True
self.on_epoch_end()
def __len__(self):
"""
Get the number of batches in the dataset.
Returns:
int: The number of batches in the dataset.
"""
return int(np.ceil(len(self.image_names) / self.batch_size))
def on_epoch_end(self):
"""
Shuffle the training set at the end of each each epoch (if shuffle == True)
"""
if self.shuffle:
self.image_names, self.mask_names = shuffle(self.image_names, self.mask_names)
def read_image_mask(self, image_name, mask_name, path):
"""
Read and preprocess an image and its corresponding mask.
Args:
image_name (str): The filename of the image.
mask_name (str): The filename of the mask.
path (str): The root path of the images and masks.
Returns:
tuple: A tuple containing the preprocessed image and mask arrays.
"""
image_path = path + '/images/'
mask_path = path + '/masks/'
image = plt.imread(os.path.join(image_path, image_name)).astype(np.uint8)
if image.shape[2] == 4:
image = image[:, :, :3]
mask = plt.imread(os.path.join(mask_path, mask_name))
mask = (mask > 0.5).astype(np.uint8)
return image, mask
def __getitem__(self, index):
"""
Generate one batch of preprocessed data for the given index.
Args:
index (int): The index of the batch.
Returns:
tuple: A tuple containing the preprocessed input (X) and target (y) arrays for the batch.
"""
start = index * self.batch_size
end = (index + 1) * self.batch_size
indexes = self.image_names[start:end]
im_in_batch = len(indexes)
X = np.zeros((im_in_batch, self.image_size, self.image_size, 3), dtype=np.float32)
y = np.zeros((im_in_batch, self.image_size, self.image_size, 1), dtype=np.float32)
for i, sample_id in enumerate(indexes):
image, mask = self.read_image_mask(self.image_names[index * self.batch_size + i],
self.mask_names[index * self.batch_size + i],
self.path)
if self.transform:
transformed = self.transform()(image=image, mask=mask)
image_trans = transformed['image']
mask_trans = transformed['mask']
if image_trans.shape[0] < 256:
image_trans = cv2.resize(image_trans, (256, 256))
mask_trans = cv2.resize(mask_trans, (256, 256))
X[i, ...] = image_trans / 255.0
y[i, ...] = np.expand_dims(mask_trans, -1)
elif not self.transform and self.batch_size == 1:
return image.reshape(1, image.shape[0], image.shape[1], 3) / 255.0, \
mask.reshape(1, mask.shape[0], mask.shape[1], 1)
return X, y
if __name__ == "__main__":
train_generator = DataGenerator(root_dir='datasets/prepared_datasets/monuseg_2018/train',
image_folder='images',
mask_folder='masks',
image_size=256,
batch_size=4,
transform=transform,
shuffle=True)
train_sample_image , train_sample_label = train_generator.__getitem__(2)
print(train_sample_image.shape , train_sample_label.shape)
val_generator = DataGenerator(root_dir='datasets/prepared_datasets/monuseg_2018/validation',
image_folder='images',
mask_folder='masks',
image_size=256,
batch_size=1,
transform=None,
shuffle=True)
val_sample_image , val_sample_label = val_generator.__getitem__(2)
print(val_sample_image.shape , val_sample_label.shape)