-
Notifications
You must be signed in to change notification settings - Fork 2
/
add_noise.py
214 lines (169 loc) · 6.62 KB
/
add_noise.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
import numpy as np
import threading
import torch
class SequentialSelect(object):
def __pos(self, n):
i = 0
while True:
yield i
i = (i + 1) % n
def __init__(self, transforms):
self.transforms = transforms
self.pos = LockedIterator(self.__pos(len(transforms)))
def __call__(self, img):
out = self.transforms[next(self.pos)](img)
return out
class LockedIterator(object):
def __init__(self, it):
self.lock = threading.Lock()
self.it = it.__iter__()
def __iter__(self):
return self
def __next__(self):
self.lock.acquire()
try:
return next(self.it)
finally:
self.lock.release()
class AddNoise(object):
"""add gaussian noise to the given numpy array (B,H,W)"""
def __init__(self, sigma):
self.sigma_ratio = sigma / 255.
def __call__(self, img):
noise = np.random.randn(*img.shape) * self.sigma_ratio
return img + noise
class AddNoiseBlind(object):
"""add blind gaussian noise to the given numpy array (B,H,W)"""
def __pos(self, n):
i = 0
while True:
yield i
i = (i + 1) % n
def __init__(self, sigmas):
self.sigmas = np.array(sigmas) / 255.
self.pos = LockedIterator(self.__pos(len(sigmas)))
def __call__(self, img):
noise = np.random.randn(*img.shape) * self.sigmas[next(self.pos)]
return img + noise
class AddNoiseBlindv2(object):
"""add blind gaussian noise to the given numpy array (B,H,W)"""
def __init__(self, min_sigma, max_sigma):
self.min_sigma = min_sigma
self.max_sigma = max_sigma
def __call__(self, img):
noise = np.random.randn(*img.shape) * np.random.uniform(
self.min_sigma, self.max_sigma) / 255
return img + noise
class AddNoiseNoniid(object):
"""add non-iid gaussian noise to the given numpy array (B,H,W)"""
def __init__(self, sigmas):
self.sigmas = np.array(sigmas) / 255.
def __call__(self, img):
bwsigmas = np.reshape(
self.sigmas[np.random.randint(0, len(self.sigmas), img.shape[0])],
(-1, 1, 1))
noise = np.random.randn(*img.shape) * bwsigmas
return img + noise
class AddNoiseMixed(object):
"""add mixed noise to the given numpy array (B,H,W)
Args:
noise_bank: list of noise maker (e.g. AddNoiseImpulse)
num_bands: list of number of band which is corrupted by each item in noise_bank"""
def __init__(self, noise_bank, num_bands):
assert len(noise_bank) == len(num_bands)
self.noise_bank = noise_bank
self.num_bands = num_bands
def __call__(self, img):
B, H, W = img.shape
all_bands = np.random.permutation(range(B))
pos = 0
for noise_maker, num_band in zip(self.noise_bank, self.num_bands):
if 0 < num_band <= 1:
num_band = int(np.floor(num_band * B))
bands = all_bands[pos:pos + num_band]
pos += num_band
img = noise_maker(img, bands)
return img
class _AddNoiseImpulse(object):
"""add impulse noise to the given numpy array (B,H,W)"""
def __init__(self, amounts, s_vs_p=0.5):
self.amounts = np.array(amounts)
self.s_vs_p = s_vs_p
def __call__(self, img, bands):
# bands = np.random.permutation(range(img.shape[0]))[:self.num_band]
bwamounts = self.amounts[np.random.randint(0, len(self.amounts),
len(bands))]
for i, amount in zip(bands, bwamounts):
self.add_noise(img[i, ...],
amount=amount,
salt_vs_pepper=self.s_vs_p)
return img
def add_noise(self, image, amount, salt_vs_pepper):
out = image
p = amount
q = salt_vs_pepper
flipped = np.random.choice([True, False],
size=image.shape,
p=[p, 1 - p])
salted = np.random.choice([True, False],
size=image.shape,
p=[q, 1 - q])
peppered = ~salted
out[torch.tensor(flipped & salted)] = 1
out[torch.tensor(flipped & peppered)] = 0
return out
class _AddNoiseStripe(object):
"""add stripe noise to the given numpy array (B,H,W)"""
def __init__(self, min_amount, max_amount):
assert max_amount > min_amount
self.min_amount = min_amount
self.max_amount = max_amount
def __call__(self, img, bands):
B, H, W = img.shape
# bands = np.random.permutation(range(img.shape[0]))[:len(bands)]
num_stripe = np.random.randint(np.floor(self.min_amount * W),
np.floor(self.max_amount * W),
len(bands))
for i, n in zip(bands, num_stripe):
loc = np.random.permutation(range(W))
loc = loc[:n]
stripe = np.random.uniform(0, 1, size=(len(loc), )) * 0.5 - 0.25
img[i, :, loc] -= np.reshape(stripe, (-1, 1))
return img
class _AddNoiseDeadline(object):
"""add deadline noise to the given numpy array (B,H,W)"""
def __init__(self, min_amount, max_amount):
assert max_amount > min_amount
self.min_amount = min_amount
self.max_amount = max_amount
def __call__(self, img, bands):
B, H, W = img.shape
# bands = np.random.permutation(range(img.shape[0]))[:len(bands)]
num_deadline = np.random.randint(np.ceil(self.min_amount * W),
np.ceil(self.max_amount * W),
len(bands))
for i, n in zip(bands, num_deadline):
loc = np.random.permutation(range(W))
loc = loc[:n]
img[i, :, loc] = 0
return img
class AddNoiseImpulse(AddNoiseMixed):
def __init__(self):
self.noise_bank = [_AddNoiseImpulse([0.1, 0.3, 0.5, 0.7])]
self.num_bands = [1 / 3]
class AddNoiseStripe(AddNoiseMixed):
def __init__(self):
self.noise_bank = [_AddNoiseStripe(0.05, 0.25)]
self.num_bands = [1 / 3]
class AddNoiseDeadline(AddNoiseMixed):
def __init__(self):
self.noise_bank = [_AddNoiseDeadline(0.05, 0.25)]
self.num_bands = [1 / 3]
class AddNoiseComplex(AddNoiseMixed):
def __init__(self):
self.noise_bank = [
_AddNoiseStripe(0.05, 0.25),
_AddNoiseDeadline(0.05, 0.25),
_AddNoiseImpulse([0.1, 0.3, 0.5, 0.7])
]
self.num_bands = [1 / 3, 1 / 3, 1 / 3]