-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathcustom_transforms.py
214 lines (163 loc) · 7.97 KB
/
custom_transforms.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
from torchvision import transforms, utils
import random
import cv2
import numpy as np
from opts import opt
import pdb
from skimage import io
class ToTensor(object):
def __init__(self):
self.to_tensor = transforms.ToTensor()
def __call__(self, sample):
if 'gt' in sample:
sample['gt'] = self.to_tensor(sample['gt'])
if 'blur' in sample:
sample['blur'] = self.to_tensor(sample['blur'])
if 'guide' in sample:
sample['guide'] = self.to_tensor(sample['guide'])
# if 'mask' in sample:
# sample['mask'] = self.to_tensor(sample['mask'])
return sample
# 图像退化方式
class GaussianBlur(object):
def __init__(self, sigma=3, size=13):
assert isinstance(sigma, (int, float))
self.sigma = sigma
assert isinstance(size, (int, tuple, list))
if isinstance(size, int):
self.size = (size, size) # size must be odd
else:
assert len(size) == 2, "len(size) of GaussianBlur must be 2!"
self.size = size
def __call__(self, sample):
if self.sigma > 0:
sample['blur'] = cv2.GaussianBlur(sample['blur'], self.size, self.sigma)
return sample
class DownSampler(object):
def __init__(self, scale):
assert isinstance(scale, (int, float))
self.scale = scale
def __call__(self, sample):
if self.scale > 1:
h, w, _ = sample['blur'].shape
scaled_h, scaled_w = int(h / self.scale), int(w / self.scale)
# downsample + upsample
sample['blur'] = cv2.resize(sample['blur'], (scaled_w, scaled_h), interpolation = cv2.INTER_CUBIC)
return sample
class UpSampler(object):
def __init__(self, scale):
assert isinstance(scale, (int, float))
self.scale = scale
def __call__(self, sample):
if self.scale > 1:
sample['blur'] = cv2.resize(sample['blur'], (opt.img_size, opt.img_size), interpolation = cv2.INTER_CUBIC)
return sample
class AWGN(object):
def __init__(self, level):
assert isinstance(level, (int, float))
self.level = level
def __call__(self, sample):
if self.level > 0:
noise = np.random.randn(*sample['blur'].shape) * self.level
# clip(0,255) 防止负数变为255
sample['blur'] = (sample['blur'] + noise).clip(0,255).astype(np.uint8) # otherwise would be np.float64
return sample
# jpeg compressor + decompressor
class JPEGCompressor(object):
def __init__(self, quality):
assert isinstance(quality, (int, float))
self.quality = quality
def __call__(self, sample):
if self.quality > 0: # 0 indicating no lossy compression (i.e losslessly compression)
encode_param = [int(cv2.IMWRITE_JPEG_QUALITY), self.quality]
sample['blur'] = cv2.imdecode(cv2.imencode('.jpg', sample['blur'], encode_param)[1], 1)
return sample
class DegradationModel(object):
def degradation_kind(self, kind = 'original'):
if kind == 'original':
# self.gaussianBlur_sigma_list = [1 + x * 0.1 for x in range(21)]
self.gaussianBlur_sigma_list = [1 + x for x in range(3)]
self.gaussianBlur_sigma_list += [0]
# self.gaussianBlur_sigma_list += int(len(self.gaussianBlur_sigma_list)) * [0] # 1/2 trigger this degradation
self.downsample_scale_list = [1 + x * 0.1 for x in range(0,71)]
# self.downsample_scale_list += int(len(self.downsample_scale_list)) * [1]
self.awgn_level_list = list(range(1, 8, 1))
# self.awgn_level_list += int(len(self.awgn_level_list)) * [0]
self.jpeg_quality_list = list(range(10, 41, 1))
self.jpeg_quality_list += int(len(self.jpeg_quality_list) * 0.33) * [0]
elif kind == 'only_downsample':
self.gaussianBlur_sigma_list = [0]
self.downsample_scale_list = [1 + x * 0.1 for x in range(0,71)]
self.awgn_level_list = [0]
self.jpeg_quality_list = [0]
elif kind == 'only_4x':
self.gaussianBlur_sigma_list = [0]
# self.downsample_scale_list = [1 + x * 0.1 for x in range(0,71)]
self.downsample_scale_list = [4]
self.awgn_level_list = [0]
self.jpeg_quality_list = [0]
elif kind == 'weaker_1': # 0.5 trigger prob
self.gaussianBlur_sigma_list = [1 + x for x in range(3)]
self.gaussianBlur_sigma_list += int(len(self.gaussianBlur_sigma_list)) * [0] # 1/2 trigger this degradation
self.downsample_scale_list = [1 + x * 0.1 for x in range(0,71)]
self.downsample_scale_list += int(len(self.downsample_scale_list)) * [1]
self.awgn_level_list = list(range(1, 8, 1))
self.awgn_level_list += int(len(self.awgn_level_list)) * [0]
self.jpeg_quality_list = list(range(10, 41, 1))
self.jpeg_quality_list += int(len(self.jpeg_quality_list)) * [0]
elif kind == 'weaker_2': # weaker than weaker_1, jpeg [20,40]
self.gaussianBlur_sigma_list = [1 + x for x in range(3)]
self.gaussianBlur_sigma_list += int(len(self.gaussianBlur_sigma_list)) * [0] # 1/2 trigger this degradation
self.downsample_scale_list = [1 + x * 0.1 for x in range(0,71)]
self.downsample_scale_list += int(len(self.downsample_scale_list)) * [1]
self.awgn_level_list = list(range(1, 8, 1))
self.awgn_level_list += int(len(self.awgn_level_list)) * [0]
self.jpeg_quality_list = list(range(20, 41, 1))
self.jpeg_quality_list += int(len(self.jpeg_quality_list)) * [0]
def __init__(self, kind = 'original', jpeg_last = False, msg = None):
# self.msg = msg
self.jpeg_last = jpeg_last
self.gaussianBlur_size_list = list(range(3,14,2))
self.degradation_kind(kind)
# ops
self.gaussianBlur = GaussianBlur(random.choice(self.gaussianBlur_sigma_list), random.choice(self.gaussianBlur_size_list))
self.downSampler = DownSampler(random.choice(self.downsample_scale_list))
self.upSampler = UpSampler(self.downSampler.scale)
self.awgn = AWGN(random.choice(self.awgn_level_list))
self.jpegCompressor = JPEGCompressor(random.choice(self.jpeg_quality_list))
def random_params(self):
self.gaussianBlur.sigma = random.choice(self.gaussianBlur_sigma_list)
self.gaussianBlur.size = (random.choice(self.gaussianBlur_size_list),) * 2
self.downSampler.scale = random.choice(self.downsample_scale_list)
self.upSampler.scale = self.downSampler.scale
self.awgn.level = random.choice(self.awgn_level_list)
self.jpegCompressor.quality = random.choice(self.jpeg_quality_list)
def __call__(self, sample):
# print (self.msg)
self.random_params()
if self.jpeg_last:
return self.jpegCompressor(self.upSampler(self.awgn(self.downSampler(self.gaussianBlur(sample)))))
else:
return self.upSampler(self.jpegCompressor(self.awgn(self.downSampler(self.gaussianBlur(sample)))))
def test_jpeg():
test_img = './sn.jpg'
gauss = GaussianBlur(1)
awgn = AWGN(100)
jpeg_d = JPEGCompressor(20)
# img = io.imread(test_img)
img = cv2.imread(test_img)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# pdb.set_trace()
img_d = awgn({'blur':img})['blur']
img_d_bgr = cv2.cvtColor(img_d, cv2.COLOR_RGB2BGR)
# pdb.set_trace()
cv2.imwrite('sn_jpeg.png', img_d_bgr)
def test_DegradationModel():
test_img = './sn.jpg'
degradationModel = DegradationModel()
img = cv2.imread(test_img)
cv2.imwrite("degraded_result.png", degradationModel({'blur':img})['blur'])
# cv2.imwrite("degraded_result.png", img)
if __name__ == '__main__':
# test_DegradationModel()
test_jpeg()