forked from jshermeyer/VDSR4Geo
-
Notifications
You must be signed in to change notification settings - Fork 8
/
data.py
224 lines (176 loc) · 9.31 KB
/
data.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
222
223
224
import os
import zipfile
import numpy as np
import math
import cv2
from tqdm import tqdm
from scipy import misc
from skimage import color
DATA_PATH = "/Set/To/Data/Path"
class TrainSet:
def __init__(self, benchmark, batch_size=64, patch_size=41, scaling_factors=(2, 4, 8)):
self.benchmark = benchmark
self.batch_size = batch_size
self.patch_size = patch_size
self.scaling_factors = scaling_factors
self.images_completed = 0
self.epochs_completed = 0
self.root_path = os.path.join(DATA_PATH, self.benchmark)
self.images = []
self.targets = []
for file_name in os.listdir(self.root_path):
#Read in image
image = misc.imread(os.path.join(self.root_path, file_name))
#Crop to an area divisible by 12
width, height = image.shape[0], image.shape[1]
width = width - width % 12
height = height - height % 12
n_horizontal_patches = width // patch_size
n_vertical_patches = height // patch_size
image= image[:width,:height]
#For each level of enhacement
for scaling_factor in scaling_factors:
#Conditional blur
blur_level=scaling_factor/2
blurred = cv2.GaussianBlur(image, (0, 0), blur_level, blur_level, 0)
#Pull out the luminance component of ycbcr for the HR and blurred images
if len(image.shape) == 3:
blurred = color.rgb2ycbcr(blurred)[:, :, 0].astype(np.uint8)
image = color.rgb2ycbcr(image)[:, :, 0].astype(np.uint8)
#downscale the blurred component
downscaled=cv2.resize(blurred, (0,0), fx=float(1 / scaling_factor),fy=float(1 / scaling_factor), interpolation=cv2.INTER_AREA)
#rescale the blurred component
rescaled = misc.imresize(downscaled, (image.shape[0],image.shape[1]), 'bicubic', mode='L')
#Save the luminance component of the original image as an HR target
high_res_image = image.astype(np.float32) / 255
#Save the blurred, downscaled/rescaled as a LR target
low_res_image = np.clip(rescaled.astype(np.float32) / 255, 0.0, 1.0)
#Create patches and data aug for training
for horizontal_patch in range(n_horizontal_patches):
for vertical_patch in range(n_vertical_patches):
h_start = horizontal_patch * patch_size
v_start = vertical_patch * patch_size
high_res_patch = high_res_image[h_start:h_start + patch_size, v_start:v_start + patch_size]
low_res_patch = low_res_image[h_start:h_start + patch_size, v_start:v_start + patch_size]
for _ in range(4):
high_res_patch = np.rot90(high_res_patch)
low_res_patch = np.rot90(low_res_patch)
self.targets.append(np.expand_dims(high_res_patch, axis=2))
self.images.append(np.expand_dims(low_res_patch, axis=2))
high_res_patch = np.fliplr(high_res_patch)
low_res_patch = np.fliplr(low_res_patch)
for _ in range(4):
high_res_patch = np.rot90(high_res_patch)
low_res_patch = np.rot90(low_res_patch)
self.targets.append(np.expand_dims(high_res_patch, axis=2))
self.images.append(np.expand_dims(low_res_patch, axis=2))
self.images = np.array(self.images)
self.targets = np.array(self.targets)
self.shuffle()
self.length = len(self.images)
self.length = self.length - self.length % batch_size
self.images = self.images[:self.length]
self.targets = self.targets[:self.length]
def batch(self):
images = self.images[self.images_completed:(self.images_completed + self.batch_size)]
targets = self.targets[self.images_completed:(self.images_completed + self.batch_size)]
self.images_completed += self.batch_size
if self.images_completed >= self.length:
self.images_completed = 0
self.epochs_completed += 1
self.shuffle()
return images, targets
def shuffle(self):
indices = list(range(len(self.images)))
np.random.shuffle(indices)
self.images = self.images[indices]
self.targets = self.targets[indices]
class TestSet:
def __init__(self, benchmark, scaling_factors=(2, 4, 8)):
self.benchmark = benchmark
self.scaling_factors = scaling_factors
self.images_completed = 0
self.root_path = os.path.join(DATA_PATH, self.benchmark)
self.file_names = os.listdir(self.root_path)
self.images = []
self.targets = []
for file_name in tqdm(os.listdir(self.root_path)):
image = misc.imread(os.path.join(self.root_path, file_name))
#For each enhancement level...
for scaling_factor in self.scaling_factors:
#Conditional Blur
blur_level=scaling_factor/2
blurred = cv2.GaussianBlur(image, (0, 0), blur_level, blur_level, 0)
if len(image.shape) == 3:
#Pull out all the original ycbcr components
ycbcr = color.rgb2ycbcr(blurred)
y = ycbcr[:, :, 0].astype(np.uint8)
b = ycbcr[:, :, 1].astype(np.uint8)
r = ycbcr[:, :, 2].astype(np.uint8)
else:
y = blurred
#Downscale them
downscaled=cv2.resize(y, (0,0), fx=float(1 / scaling_factor),fy=float(1 / scaling_factor), interpolation=cv2.INTER_AREA)
d_b=cv2.resize(b, (0,0), fx=float(1 / scaling_factor),fy=float(1 / scaling_factor), interpolation=cv2.INTER_AREA)
d_r=cv2.resize(r, (0,0), fx=float(1 / scaling_factor),fy=float(1 / scaling_factor), interpolation=cv2.INTER_AREA)
#rescale them
rescaled = misc.imresize(downscaled, (y.shape[0],y.shape[1]), 'bicubic', mode='L')
r_b = misc.imresize(d_b, (y.shape[0],y.shape[1]), 'bicubic', mode='L')
d_r = misc.imresize(d_r, (y.shape[0],y.shape[1]), 'bicubic', mode='L')
#Create the LR image to convert to HR
if len(image.shape) == 3:
low_res_image = ycbcr
low_res_image[:, :, 0] = rescaled
low_res_image[:, :, 1] = r_b
low_res_image[:, :, 2] = d_r
low_res_image = color.ycbcr2rgb(low_res_image)
low_res_image = (np.clip(low_res_image, 0.0, 1.0) * 255).astype(np.uint8)
else:
low_res_image = rescaled
self.images.append(low_res_image)
self.targets.append(image)
self.length = len(self.images)
def fetch(self):
if self.images_completed >= self.length:
return None
else:
self.images_completed += 1
return self.images[self.images_completed - 1], self.targets[self.images_completed - 1]
class SR_Run:
def __init__(self, benchmark, scaling_factors=(2, 4, 8)):
self.benchmark = benchmark
self.scaling_factors = scaling_factors
self.images_completed = 0
self.root_path = os.path.join(DATA_PATH, self.benchmark)
self.file_names = os.listdir(self.root_path)
self.images = []
self.targets = []
for file_name in tqdm(os.listdir(self.root_path)):
image = misc.imread(os.path.join(self.root_path, file_name))
for scaling_factor in self.scaling_factors:
if len(image.shape) == 3:
ycbcr = color.rgb2ycbcr(image)
downscaled = ycbcr[:, :, 0].astype(np.uint8)
d_b = ycbcr[:, :, 1].astype(np.uint8)
d_r = ycbcr[:, :, 2].astype(np.uint8)
else:
y = image
rescaled = misc.imresize(downscaled, float(scaling_factor), 'bicubic', mode='L')
r_b = misc.imresize(d_b, float(scaling_factor), 'bicubic', mode='L')
d_r = misc.imresize(d_r, float(scaling_factor), 'bicubic', mode='L')
if len(image.shape) == 3:
low_res_image = np.stack([rescaled,r_b,d_r],axis=2)
low_res_image=low_res_image.astype(np.float64)
low_res_image = color.ycbcr2rgb(low_res_image)
low_res_image = (np.clip(low_res_image, 0.0, 1.0) * 255).astype(np.uint8)
else:
low_res_image = rescaled
self.images.append(low_res_image)
self.targets.append(image)
self.length = len(self.images)
def fetch(self):
if self.images_completed >= self.length:
return None
else:
self.images_completed += 1
return self.images[self.images_completed - 1], self.targets[self.images_completed - 1]