-
Notifications
You must be signed in to change notification settings - Fork 0
/
ndarray_augmentor.py
195 lines (164 loc) · 6.92 KB
/
ndarray_augmentor.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
import cv2
import numpy as np
import os
import uuid
import copy
def rotateImage(image, angle):
l = len(image.shape)
image_center = tuple(np.array(image.shape[:2]) / 2)
rot_mat = cv2.getRotationMatrix2D(image_center, angle, 1.0)
result = cv2.warpAffine(image, rot_mat, image.shape[:2], flags=cv2.INTER_LINEAR)
if len(result.shape) < l:
y, x = result.shape
result = result.reshape((y, x, 1))
return result
def zoom(image, zoom_scale):
size = image.shape
l = len(size)
image = cv2.resize(image, None, fx=zoom_scale, fy=zoom_scale)
if len(image.shape) < l:
y, x = image.shape
image = image.reshape((y, x, 1))
new_size = image.shape
if len(size) == 3:
if zoom_scale > 1:
return image[int((new_size[0] - size[0]) / 2): int((new_size[0] - size[0]) / 2 + size[0]),
int((new_size[1] - size[1]) / 2): int((new_size[1] - size[1]) / 2 + size[1]), :]
elif zoom_scale == 1:
return image
else:
new_image = np.zeros(size).astype('uint8')
new_image[int((size[0] - new_size[0]) / 2): int((size[0] - new_size[0]) / 2 + new_size[0]),
int((size[1] - new_size[1]) / 2): int((size[1] - new_size[1]) / 2 + new_size[1]), :] = image
return new_image
def sample(image, rotation_min, rotation_max, fliplr, flipud, zoom_min, zoom_max):
angle = np.random.uniform(rotation_min, rotation_max)
image = rotateImage(image, angle)
if fliplr:
if np.random.random() < 0.5:
image = np.fliplr(image)
if flipud:
if np.random.random() < 0.5:
image = np.flipud(image)
zoom_scale = np.random.uniform(zoom_min, zoom_max)
image = zoom(image, zoom_scale)
return image
def augmentor(npy_path, save_path, batch_number=1, rotation_min=0, rotation_max=0, fliplr=False, flipud=False, zoom_min=1, zoom_max=1, input_c=6):
c = 0
npy_files = [os.path.join(npy_path, f) for f in os.listdir(npy_path) if f.endswith('.npy')]
while c < batch_number:
for npy_file in npy_files:
data = np.load(npy_file)
data = sample(data, rotation_min, rotation_max, fliplr, flipud, zoom_min, zoom_max)
unid = uuid.uuid4().hex
image = data[:, :, :input_c]
masks = data[:, :, input_c:]
num_objs = masks.shape[2]
print(masks.shape[2])
for i in reversed(range(num_objs)):
mask = masks[:, :, i]
if mask.max() < 250:
masks = np.delete(masks, i, axis=2)
continue
mask = mask >= 250
pos = np.where(mask)
xmin = np.min(pos[1])
xmax = np.max(pos[1])
ymin = np.min(pos[0])
ymax = np.max(pos[0])
if xmin >= xmax:
masks = np.delete(masks, i, axis=2)
continue
if ymin >= ymax:
masks = np.delete(masks, i, axis=2)
continue
print(masks.shape[2])
print('\n')
data = np.append(image, masks, axis=2)
save_file = npy_file.split('.npy')[0].split('/')[-1] + "_" + unid + ".npy"
save_file = os.path.join(save_path, save_file)
np.save(save_file, data)
c += 1
def balanced_augmentor(image_path, label_path, aug_path, augmentation_batch=1, augmentation_ratio=[],
rotation_min=0, rotation_max=0, fliplr=False, flipud=False, zoom_min=1, zoom_max=1):
image_files = [os.path.join(image_path, f) for f in os.listdir(image_path) if f.endswith('.jpg')]
while augmentation_batch:
augmentation_batch -= 1
for image_file in image_files:
f_name = image_file.split('/')[-1][:-4]
mask_file = os.path.join(label_path, f_name + "_nd.npy")
cls_file = os.path.join(label_path, f_name + "_cls.npy")
if not os.path.isfile(mask_file):
continue
if not os.path.isfile(cls_file):
continue
image = cv2.imread(image_file)
masks = np.load(mask_file)
clses = np.load(cls_file) # cls should start with 0, e.g. [0, 1, 2, 3, ...]
image_mask = np.concatenate((image, masks), axis=2)
n = augmentation_ratio[int(clses.max())]
for _ in range(n):
data = copy.deepcopy(image_mask)
data = sample(data, rotation_min, rotation_max, fliplr, flipud, zoom_min, zoom_max)
image = data[:, :, :3]
masks = data[:, :, 3:]
cls = copy.deepcopy(clses)
if masks.max() < 0.1:
continue
print(masks.shape[2])
num_objs = masks.shape[2]
for i in reversed(range(num_objs)):
mask = masks[:, :, i]
if mask.max() < 250:
masks = np.delete(masks, i, axis=2)
cls = np.delete(cls, i)
continue
mask = mask >= 250
pos = np.where(mask)
xmin = np.min(pos[1])
xmax = np.max(pos[1])
ymin = np.min(pos[0])
ymax = np.max(pos[0])
if (xmin >= xmax) | (ymin >= ymax):
masks = np.delete(masks, i, axis=2)
cls = np.delete(cls, i)
continue
if masks.shape[2] == 0:
continue
if masks.max() == 0:
continue
print(masks.shape[2])
print('\n')
unid = uuid.uuid4().hex
new_mask_file = os.path.join(aug_path, f_name + "_" + unid + "_nd.npy")
new_cls_file = os.path.join(aug_path, f_name + "_" + unid + "_cls.npy")
new_image_file = os.path.join(aug_path, f_name + "_" + unid + ".jpg")
np.save(new_mask_file, masks)
np.save(new_cls_file, cls)
cv2.imwrite(new_image_file, image)
if __name__ == '__main__':
config = dict(
batch_number=6,
rotation_min=0,#-90,
rotation_max=0,#90,
fliplr=True,
flipud=True,
zoom_min=1,#0.8,
zoom_max=1,#1.2,
input_c=6)
config_ = dict(
augmentation_batch=5,
augmentation_ratio=[1, 3, 15, 50, 100],
rotation_min=-90,
rotation_max=90,
fliplr=True,
flipud=True,
zoom_min=0.8,
zoom_max=1.2)
#image_path = './datasets/Eureka/images/'
#label_path = './datasets/Eureka/labels/'
#aug_path = './datasets/Eureka/aug/'
#balanced_augmentor(image_path, label_path, aug_path, **config_)
npy_path = './datasets/hypolith/rgb_masks/'
aug_path = './datasets/hypolith/aug/'
augmentor(npy_path, aug_path, **config)