-
Notifications
You must be signed in to change notification settings - Fork 1
/
dataset.py
195 lines (165 loc) · 7.77 KB
/
dataset.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 torch.utils.data as data
import torch
import glob
import numpy as np
import torchvision.transforms as transforms
import tifffile as tiff
import scipy.misc as smi
import cv2
import time
from PIL import Image
import scipy.io as scio
def get_lrhsi(img, degradation_mode):
# print("degradation_mode:", degradation_mode)
if degradation_mode == 0:
scale_factor = 8
dim = np.shape(img)
img_down = np.zeros([dim[0], int(dim[1]/scale_factor), int(dim[2]/scale_factor)])
img_rebuild = np.zeros(dim)
# gaussian blur kernel, size:8x8, standrad deviation:3
kernel = np.array([[0.0067, 0.0094, 0.0118, 0.0131, 0.0131, 0.0118, 0.0094, 0.0067],
[0.0094, 0.0131, 0.0164, 0.0183, 0.0183, 0.0164, 0.0131, 0.0094],
[0.0118, 0.0164, 0.0205, 0.0229, 0.0229, 0.0205, 0.0164, 0.0118],
[0.0131, 0.0183, 0.0229, 0.0256, 0.0256, 0.0229, 0.0183, 0.0131],
[0.0131, 0.0183, 0.0229, 0.0256, 0.0256, 0.0229, 0.0183, 0.0131],
[0.0118, 0.0164, 0.0205, 0.0229, 0.0229, 0.0205, 0.0164, 0.0118],
[0.0094, 0.0131, 0.0164, 0.0183, 0.0183, 0.0164, 0.0131, 0.0094],
[0.0067, 0.0094, 0.0118, 0.0131, 0.0131, 0.0118, 0.0094, 0.0067]])
kernel = [kernel] * dim[0]
for i in range(int(dim[1]/scale_factor)):
for j in range(int(dim[2]/scale_factor)):
img_down[:, i, j] = np.sum(np.sum(img[:, i*scale_factor:(i+1)*scale_factor, j*scale_factor:(j+1)*scale_factor] * kernel, axis=1), axis=1)
elif degradation_mode == 1:
scale_factor = 8
dim = np.shape(img)
img_down = np.zeros([dim[0], int(dim[1]/scale_factor), int(dim[2]/scale_factor)])
img_rebuild = np.zeros(dim)
# uniform blur kernel, size: scale_factor * scale_factor
kernel = np.ones([int(scale_factor), int(scale_factor)]) / (scale_factor**2)
kernel = [kernel] * dim[0]
for i in range(int(dim[1]/scale_factor)):
for j in range(int(dim[2]/scale_factor)):
img_down[:, i, j] = np.sum(np.sum(img[:, i*scale_factor:(i+1)*scale_factor, j*scale_factor:(j+1)*scale_factor] * kernel, axis=1), axis=1)
elif degradation_mode == 2:
scale_factor = 16
dim = np.shape(img)
img_down = np.zeros([dim[0], int(dim[1]/scale_factor), int(dim[2]/scale_factor)])
img_rebuild = np.zeros(dim)
# uniform blur kernel, size: scale_factor * scale_factor
kernel = np.ones([int(scale_factor), int(scale_factor)]) / (scale_factor**2)
kernel = [kernel] * dim[0]
for i in range(int(dim[1]/scale_factor)):
for j in range(int(dim[2]/scale_factor)):
img_down[:, i, j] = np.sum(np.sum(img[:, i*scale_factor:(i+1)*scale_factor, j*scale_factor:(j+1)*scale_factor] * kernel, axis=1), axis=1)
else:
scale_factor = 32
dim = np.shape(img)
img_down = np.zeros([dim[0], int(dim[1]/scale_factor), int(dim[2]/scale_factor)])
img_rebuild = np.zeros(dim)
# uniform blur kernel, size: scale_factor * scale_factor
kernel = np.ones([int(scale_factor), int(scale_factor)]) / (scale_factor**2)
kernel = [kernel] * dim[0]
for i in range(int(dim[1]/scale_factor)):
for j in range(int(dim[2]/scale_factor)):
img_down[:, i, j] = np.sum(np.sum(img[:, i*scale_factor:(i+1)*scale_factor, j*scale_factor:(j+1)*scale_factor] * kernel, axis=1), axis=1)
for i in range(dim[0]):
img_down_slice = Image.fromarray(img_down[i, :, :])
img_rebuild[i, :, :] = img_down_slice.resize((dim[1],dim[2]), Image.BICUBIC)
# img_rebuild[i, :, :] = cv2.resize(img_down[i, :, :],(dim[1],dim[2]),interpolation=cv2.INTER_LINEAR)
return img_rebuild.astype(np.float32)
def flip(img, random_flip):
# input:3D array to flip
# output:3D array flipped
channel = np.shape(img)[0]
# Random horizontal flipping
if random_flip[0] > 0.5:
for i in range(channel):
img[i, :, :] = np.fliplr(img[i, :, :])
# Random vertical flipping
if random_flip[1] > 0.5:
for i in range(channel):
img[i, :, :] = np.flipud(img[i, :, :])
# Rotate img by 90 degrees
if random_flip[2] > 0.5:
for i in range(channel):
img[i, :, :] = np.rot90(img[i, :, :])
return img
class Dataset_cave_train(data.Dataset):
def __init__(self, file_train_path):
self.tif_list = glob.glob(file_train_path+'/*.tif')
self.transforms = transforms.ToTensor()
# self.scale_factor = scale_factor
def __getitem__(self, index):
index = index % len(self.tif_list)
img = tiff.imread(self.tif_list[index])
# print(int(time.time()*1000))
np.random.seed(int(time.time()*1000)%1000000)
topleft = np.random.random_integers(0, 512-128, 2)
degradation_mode = np.random.random_integers(0, 4)
img = img[:, topleft[0]:topleft[0]+128, topleft[1]:topleft[1]+128]
random_flip = np.random.random(3)
img = flip(img, random_flip)
img_hsi = img[0:31, :, :]
img_rgb = img[31:31+3, :, :]
img_hsi_rebuild = get_lrhsi(img_hsi, degradation_mode)
data_hsi = self.transforms(img_hsi_rebuild.transpose([1, 2, 0]))
data_rgb = self.transforms(img_rgb.transpose([1, 2, 0]))
label = self.transforms(img_hsi.transpose([1, 2, 0]))
return data_hsi, data_rgb, label
def __len__(self):
return 100 * len(self.tif_list) # num of patchs is 100 in one raw train image
class Dataset_cave_val(data.Dataset):
def __init__(self, file_test_path):
self.tif_list = glob.glob(file_test_path+'/*.tif')
self.transforms = transforms.ToTensor()
# self.scale_factor = scale_factor
def __getitem__(self, index):
index = index % len(self.tif_list)
img = tiff.imread(self.tif_list[index])
np.random.seed(int(time.time()*1000)%1000000)
topleft = np.random.random_integers(0, 512-128, 2)
degradation_mode = np.random.random_integers(0, 4)
img = img[:, topleft[0]:topleft[0]+128, topleft[1]:topleft[1]+128]
img_hsi = img[0:31, :, :]
img_rgb = img[31:31+3, :, :]
img_hsi_rebuild = get_lrhsi(img_hsi, degradation_mode)
# img_rgb = np.random.random_integers(256, size=[3, 64, 64]).astype(np.uint8)
data_hsi = self.transforms(img_hsi_rebuild.transpose([1, 2, 0]))
data_rgb = self.transforms(img_rgb.transpose([1, 2, 0]))
label = self.transforms(img_hsi.transpose([1, 2, 0]))
return data_hsi, data_rgb, label
def __len__(self):
return 5 * len(self.tif_list)
class Dataset_cave_test(data.Dataset):
def __init__(self, file_test_path, degradation_mode):
self.tif_list = glob.glob(file_test_path+'/*.tif')
self.transforms = transforms.ToTensor()
self.degradation_mode = degradation_mode
def __getitem__(self, index):
img = tiff.imread(self.tif_list[index])
img_hsi = img[0:31, :, :]
img_rgb = img[31:31+3, :, :]
img_hsi_rebuild = get_lrhsi(img_hsi, self.degradation_mode)
# img_rgb = np.random.random_integers(256, size=[3, 512, 512]).astype(np.uint8)
data_hsi = self.transforms(img_hsi_rebuild.transpose([1, 2, 0]))
data_rgb = self.transforms(img_rgb.transpose([1, 2, 0]))
label = self.transforms(img_hsi.transpose([1, 2, 0]))
return data_hsi, data_rgb, label
def __len__(self):
return len(self.tif_list)
# # train
# file = './data/train/'
# a = Dataset_cave_train(file)
# hsi, rgb, lb = a.__getitem__(313)
# # # test
# # file = './data/test/'
# # a = Dataset_cave_test(file)
# # hsi, rgb, lb = a.__getitem__(1)
# print(lb.dtype)
# print(hsi.size())
# hsi = hsi.numpy()*255
# tiff.imsave('./data/hsi.tif', hsi.astype(np.uint8))
# lb = lb.numpy()*255
# tiff.imsave('./data/lb.tif', lb.astype(np.uint8))
# rgb = rgb.numpy()*255
# tiff.imsave('./data/rgb.tif', rgb.astype(np.uint8))