-
Notifications
You must be signed in to change notification settings - Fork 17
/
dataloader_TransMEF.py
201 lines (165 loc) · 7.53 KB
/
dataloader_TransMEF.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
# -*- coding: utf-8 -*-
# Citation:
# @article{qu2021transmef,
# title={TransMEF: A Transformer-Based Multi-Exposure Image Fusion Framework using Self-Supervised Multi-Task Learning},
# author={Qu, Linhao and Liu, Shaolei and Wang, Manning and Song, Zhijian},
# journal={arXiv preprint arXiv:2112.01030},
# year={2021}
# }
from __future__ import print_function
import torch.utils.data as Data
import torchvision.transforms as transforms
import numpy as np
from glob import glob
import os
import copy
from PIL import Image
import random
from imgaug import augmenters as iaa
sometimes = lambda aug: iaa.Sometimes(0.8, aug)
np.random.seed(2)
def local_pixel_shuffling(x):
image_temp = copy.deepcopy(x)
orig_image = copy.deepcopy(x)
img_rows, img_cols = x.shape
num_block = 10
for _ in range(num_block):
block_noise_size_x = random.randint(1, img_rows // 10)
block_noise_size_y = random.randint(1, img_cols // 10)
noise_x = random.randint(0, img_rows - block_noise_size_x)
noise_y = random.randint(0, img_cols - block_noise_size_y)
window = orig_image[noise_x:noise_x + block_noise_size_x,
noise_y:noise_y + block_noise_size_y]
window = window.flatten()
np.random.shuffle(window)
window = window.reshape((block_noise_size_x,
block_noise_size_y))
image_temp[noise_x:noise_x + block_noise_size_x,
noise_y:noise_y + block_noise_size_y] = window
local_shuffling_x = image_temp
return local_shuffling_x
def global_patch_shuffling(x):
image_temp = copy.deepcopy(x)
orig_image = copy.deepcopy(x)
img_rows, img_cols = x.shape
num_block = 10
for _ in range(num_block):
block_noise_size_x = random.randint(1, img_rows // 10)
block_noise_size_y = random.randint(1, img_cols // 10)
noise_x1 = random.randint(0, img_rows - block_noise_size_x)
noise_y1 = random.randint(0, img_cols - block_noise_size_y)
noise_x2 = random.randint(0, img_rows - block_noise_size_x)
noise_y2 = random.randint(0, img_cols - block_noise_size_y)
window1 = orig_image[noise_x1:noise_x1 + block_noise_size_x,
noise_y1:noise_y1 + block_noise_size_y]
window2 = orig_image[noise_x2:noise_x2 + block_noise_size_x,
noise_y2:noise_y2 + block_noise_size_y]
window_tmp = window1
window1 = window2
window2 = window_tmp
image_temp[noise_x1:noise_x1 + block_noise_size_x,
noise_y1:noise_y1 + block_noise_size_y] = window1
image_temp[noise_x2:noise_x2 + block_noise_size_x,
noise_y2:noise_y2 + block_noise_size_y] = window2
local_shuffling_x = image_temp
return local_shuffling_x
def brightness_aug(x, gamma):
aug_brightness = iaa.Sequential(sometimes(iaa.GammaContrast(gamma=gamma)))
aug_image = aug_brightness(images=x)
return aug_image
def bright_transform(x):
image_temp = copy.deepcopy(x)
orig_image = copy.deepcopy(x)
img_rows, img_cols = x.shape
num_block = 10
for _ in range(num_block):
block_noise_size_x = random.randint(1, img_rows // 10)
block_noise_size_y = random.randint(1, img_cols // 10)
noise_x = random.randint(0, img_rows - block_noise_size_x)
noise_y = random.randint(0, img_cols - block_noise_size_y)
window = orig_image[noise_x:noise_x + block_noise_size_x,
noise_y:noise_y + block_noise_size_y]
window = brightness_aug(window, 3 * np.random.random_sample())
image_temp[noise_x:noise_x + block_noise_size_x,
noise_y:noise_y + block_noise_size_y] = window
bright_transform_x = image_temp
return bright_transform_x
def fourier_broken(x, nb_rows, nb_cols):
aug_a = iaa.GaussianBlur(sigma=0.5)
aug_p = iaa.Jigsaw(nb_rows=nb_rows, nb_cols=nb_cols, max_steps=(1, 5))
fre = np.fft.fft2(x)
fre_a = np.abs(fre)
fre_p = np.angle(fre)
fre_a_normalize = fre_a / (np.max(fre_a) + 0.0001)
fre_p_normalize = fre_p
fre_a_trans = aug_a(image=fre_a_normalize)
fre_p_trans = aug_p(image=fre_p_normalize)
fre_a_trans = fre_a_trans * (np.max(fre_a) + 0.0001)
fre_p_trans = fre_p_trans
fre_recon = fre_a_trans * np.e ** (1j * (fre_p_trans))
img_recon = np.abs(np.fft.ifft2(fre_recon))
return img_recon
def fourier_transform(x):
image_temp = copy.deepcopy(x)
orig_image = copy.deepcopy(x)
img_rows, img_cols = x.shape
num_block = 10
for _ in range(num_block):
block_noise_size_x = random.randint(1, img_rows // 10)
block_noise_size_y = random.randint(1, img_cols // 10)
noise_x = random.randint(0, img_rows - block_noise_size_x)
noise_y = random.randint(0, img_cols - block_noise_size_y)
window = orig_image[noise_x:noise_x + block_noise_size_x,
noise_y:noise_y + block_noise_size_y]
window = fourier_broken(window, block_noise_size_x, block_noise_size_y)
image_temp[noise_x:noise_x + block_noise_size_x,
noise_y:noise_y + block_noise_size_y] = window
bright_transform_x = image_temp
return bright_transform_x
class Fusionset(Data.Dataset):
def __init__(self, io, args, root, transform=None, gray=True, partition='train', ssl_transformations=None):
self.files = glob(os.path.join(root, '*.*'))
self.gray = gray
self._tensor = transforms.ToTensor()
self.transform = transform
self.ssl_transformations = ssl_transformations
self.args = args
if args.miniset == True:
self.files = random.sample(self.files, int(args.minirate * len(self.files)))
self.num_examples = len(self.files)
if self.ssl_transformations == True:
print('used ssl_transformations')
else:
print('not used ssl_transformations')
if partition == 'train':
self.train_ind = np.asarray([i for i in range(self.num_examples) if i % 10 < 8]).astype(np.int)
np.random.shuffle(self.train_ind)
self.val_ind = np.asarray([i for i in range(self.num_examples) if i % 10 >= 8]).astype(np.int)
np.random.shuffle(self.val_ind)
io.cprint("number of " + partition + " examples in dataset" + ": " + str(len(self.files)))
def __len__(self):
return len(self.files)
def __getitem__(self, index):
img = Image.open(self.files[index])
if self.transform is not None:
img = self.transform(img)
if self.gray:
img = img.convert('L')
img = np.array(img)
if self.ssl_transformations == True:
img_bright_orig = img.copy()
img_bright_trans = bright_transform(img_bright_orig)
img_bright_trans = self._tensor(img_bright_trans)
img_fourier_orig = img.copy()
img_fourier_trans = fourier_transform(img_fourier_orig)
img_fourier_trans = self._tensor(img_fourier_trans)
img_shuffling_orig = img.copy()
img_shuffling_trans = global_patch_shuffling(img_shuffling_orig)
img_shuffling_trans = self._tensor(img_shuffling_trans)
img = self._tensor(img)
else:
img = self._tensor(img)
img_bright_trans = img
img_fourier_trans = img
img_shuffling_trans = img
return img, img_bright_trans, img_fourier_trans, img_shuffling_trans