-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathdataset_all.py
223 lines (179 loc) · 7.47 KB
/
dataset_all.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
215
216
217
218
219
220
221
import os.path
import torch
import torch.utils.data as data
from PIL import Image
import random
from random import randrange
from torchvision.transforms import ToTensor
import torchvision.transforms as transforms
IMG_EXTENSIONS = [
'.jpg', '.JPG', '.jpeg', '.JPEG',
'.png', '.PNG', '.ppm', '.PPM', '.bmp', '.BMP',
]
def is_image_file(filename):
return any(filename.endswith(extension) for extension in IMG_EXTENSIONS)
def make_dataset(dir):
images = []
assert os.path.isdir(dir), '%s is not a valid directory' % dir
for root, _, fnames in sorted(os.walk(dir)):
for fname in fnames:
if is_image_file(fname):
path = os.path.join(root, fname)
images.append(path)
return images
def rotate(img,rotate_index):
'''
:return: 8 version of rotating image
'''
if rotate_index == 0:
return img
if rotate_index==1:
return img.rotate(90)
if rotate_index==2:
return img.rotate(180)
if rotate_index==3:
return img.rotate(270)
if rotate_index==4:
return img.transpose(Image.FLIP_TOP_BOTTOM)
if rotate_index==5:
return img.rotate(90).transpose(Image.FLIP_TOP_BOTTOM)
if rotate_index==6:
return img.rotate(180).transpose(Image.FLIP_TOP_BOTTOM)
if rotate_index==7:
return img.rotate(270).transpose(Image.FLIP_TOP_BOTTOM)
class TrainLabeled(data.Dataset):
def __init__(self, dataroot, phase, finesize):
super().__init__()
self.phase = phase
self.root = dataroot
self.fineSize = finesize
self.dir_A = os.path.join(self.root, self.phase + '/input')
self.dir_B = os.path.join(self.root, self.phase + '/GT')
self.dir_C = os.path.join(self.root, self.phase + '/LA')
# image path
self.A_paths = sorted(make_dataset(self.dir_A))
self.B_paths = sorted(make_dataset(self.dir_B))
self.C_paths = sorted(make_dataset(self.dir_C))
# transform
self.transform = ToTensor() # [0,1]
def __getitem__(self, index):
# A, B is the image pair, hazy, gt respectively
A = Image.open(self.A_paths[index]).convert("RGB")
B = Image.open(self.B_paths[index]).convert("RGB")
C = Image.open(self.C_paths[index]).convert("RGB")
# resize
resized_a = A.resize((280, 280), Image.ANTIALIAS)
resized_b = B.resize((280, 280), Image.ANTIALIAS)
resized_c = C.resize((280, 280), Image.ANTIALIAS)
# crop the training image into fineSize
w, h = resized_a.size
x, y = randrange(w - self.fineSize + 1), randrange(h - self.fineSize + 1)
cropped_a = resized_a.crop((x, y, x + self.fineSize, y + self.fineSize))
cropped_b = resized_b.crop((x, y, x + self.fineSize, y + self.fineSize))
cropped_c = resized_c.crop((x, y, x + self.fineSize, y + self.fineSize))
# rotate
rotate_index = randrange(0, 8)
rotated_a = rotate(cropped_a, rotate_index)
rotated_b = rotate(cropped_b, rotate_index)
rotated_c = rotate(cropped_c, rotate_index)
# transform to (0, 1)
tensor_a = self.transform(rotated_a)
tensor_b = self.transform(rotated_b)
tensor_c = self.transform(rotated_c)
return tensor_a, tensor_b, tensor_c
def __len__(self):
return len(self.A_paths)
class TrainUnlabeled(data.Dataset):
def __init__(self, dataroot, phase, finesize):
super().__init__()
self.phase = phase
self.root = dataroot
self.fineSize = finesize
self.dir_A = os.path.join(self.root, self.phase + '/input')
self.dir_C = os.path.join(self.root, self.phase + '/LA')
self.dir_D = os.path.join(self.root, self.phase + '/candidate')
# image path
self.A_paths = sorted(make_dataset(self.dir_A))
self.C_paths = sorted(make_dataset(self.dir_C))
self.D_paths = sorted(make_dataset(self.dir_D))
# transform
self.transform = ToTensor() # [0,1]
def __getitem__(self, index):
A = Image.open(self.A_paths[index]).convert("RGB")
C = Image.open(self.C_paths[index]).convert("RGB")
candidate = Image.open(self.D_paths[index]).convert('RGB')
A = A.resize((self.fineSize, self.fineSize), Image.ANTIALIAS)
C = C.resize((self.fineSize, self.fineSize), Image.ANTIALIAS)
# strong augmentation
strong_data = data_aug(A)
tensor_w = self.transform(A)
tensor_s = self.transform(strong_data)
tensor_c = self.transform(C)
tensor_d = self.transform(candidate)
name = self.D_paths[index]
return tensor_w, tensor_s, tensor_c, tensor_d, name
def __len__(self):
return len(self.A_paths)
class ValLabeled(data.Dataset):
def __init__(self, dataroot, phase, finesize):
super().__init__()
self.phase = phase
self.root = dataroot
self.fineSize = finesize
self.dir_A = os.path.join(self.root, self.phase + '/input')
self.dir_B = os.path.join(self.root, self.phase + '/GT')
self.dir_C = os.path.join(self.root, self.phase + '/LA')
# image path
self.A_paths = sorted(make_dataset(self.dir_A))
self.B_paths = sorted(make_dataset(self.dir_B))
self.C_paths = sorted(make_dataset(self.dir_C))
# transform
self.transform = ToTensor() # [0,1]
def __getitem__(self, index):
# A, B is the image pair, hazy, gt respectively
A = Image.open(self.A_paths[index]).convert("RGB")
B = Image.open(self.B_paths[index]).convert("RGB")
C = Image.open(self.C_paths[index]).convert("RGB")
resized_a = A.resize((self.fineSize, self.fineSize), Image.ANTIALIAS)
resized_b = B.resize((self.fineSize, self.fineSize), Image.ANTIALIAS)
resized_c = C.resize((self.fineSize, self.fineSize), Image.ANTIALIAS)
# transform to (0, 1)
tensor_a = self.transform(resized_a)
tensor_b = self.transform(resized_b)
tensor_c = self.transform(resized_c)
return tensor_a, tensor_b, tensor_c
def __len__(self):
return len(self.A_paths)
class TestData(data.Dataset):
def __init__(self, dataroot):
super().__init__()
self.root = dataroot
self.dir_A = os.path.join(self.root + '/input')
self.dir_C = os.path.join(self.root + '/LA')
# image path
self.A_paths = sorted(make_dataset(self.dir_A))
self.C_paths = sorted(make_dataset(self.dir_C))
# transform
self.transform = ToTensor() # [0,1]
def __getitem__(self, index):
# A, B is the image pair, hazy, gt respectively
A = Image.open(self.A_paths[index]).convert("RGB")
C = Image.open(self.C_paths[index]).convert("RGB")
# transform to (0, 1)
tensor_a = self.transform(A)
tensor_c = self.transform(C)
return tensor_a, tensor_c
def __len__(self):
return len(self.A_paths)
def data_aug(images):
kernel_size = int(random.random() * 4.95)
kernel_size = kernel_size + 1 if kernel_size % 2 == 0 else kernel_size
blurring_image = transforms.GaussianBlur(kernel_size, sigma=(0.1, 2.0))
color_jitter = transforms.ColorJitter(brightness=0.5, contrast=0.5, saturation=0.5, hue=0.25)
strong_aug = images
if random.random() < 0.8:
strong_aug = color_jitter(strong_aug)
strong_aug = transforms.RandomGrayscale(p=0.2)(strong_aug)
if random.random() < 0.5:
strong_aug = blurring_image(strong_aug)
return strong_aug