-
Notifications
You must be signed in to change notification settings - Fork 9
/
data.py
264 lines (224 loc) · 10.6 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
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
import os
from PIL import Image
import torch.utils.data as data
import torchvision.transforms as transforms
import random
import numpy as np
from PIL import ImageEnhance
# several data augumentation strategies
def cv_random_flip(img, label, depth,edge):
flip_flag = random.randint(0, 1)
# flip_flag2= random.randint(0,1)
# left right flip
if flip_flag == 1:
img = img.transpose(Image.FLIP_LEFT_RIGHT)
label = label.transpose(Image.FLIP_LEFT_RIGHT)
depth = depth.transpose(Image.FLIP_LEFT_RIGHT)
edge = edge.transpose(Image.FLIP_LEFT_RIGHT)
# top bottom flip
# if flip_flag2==1:
# img = img.transpose(Image.FLIP_TOP_BOTTOM)
# label = label.transpose(Image.FLIP_TOP_BOTTOM)
# depth = depth.transpose(Image.FLIP_TOP_BOTTOM)
return img, label, depth, edge
def randomCrop(image, label, depth, edge):
border = 30
image_width = image.size[0]
image_height = image.size[1]
crop_win_width = np.random.randint(image_width - border, image_width)
crop_win_height = np.random.randint(image_height - border, image_height)
random_region = (
(image_width - crop_win_width) >> 1, (image_height - crop_win_height) >> 1, (image_width + crop_win_width) >> 1,
(image_height + crop_win_height) >> 1)
return image.crop(random_region), label.crop(random_region), depth.crop(random_region), edge.crop(random_region)
def randomRotation(image, label, depth, edge):
mode = Image.BICUBIC
if random.random() > 0.8:
random_angle = np.random.randint(-15, 15)
image = image.rotate(random_angle, mode)
label = label.rotate(random_angle, mode)
depth = depth.rotate(random_angle, mode)
edge = edge.rotate(random_angle, mode)
return image, label, depth, edge
def colorEnhance(image):
bright_intensity = random.randint(5, 15) / 10.0
image = ImageEnhance.Brightness(image).enhance(bright_intensity)
contrast_intensity = random.randint(5, 15) / 10.0
image = ImageEnhance.Contrast(image).enhance(contrast_intensity)
color_intensity = random.randint(0, 20) / 10.0
image = ImageEnhance.Color(image).enhance(color_intensity)
sharp_intensity = random.randint(0, 30) / 10.0
image = ImageEnhance.Sharpness(image).enhance(sharp_intensity)
return image
def randomGaussian(image, mean=0.1, sigma=0.35):
def gaussianNoisy(im, mean=mean, sigma=sigma):
for _i in range(len(im)):
im[_i] += random.gauss(mean, sigma)
return im
img = np.asarray(image)
width, height = img.shape
img = gaussianNoisy(img[:].flatten(), mean, sigma)
img = img.reshape([width, height])
return Image.fromarray(np.uint8(img))
def randomPeper(img):
img = np.array(img)
noiseNum = int(0.0015 * img.shape[0] * img.shape[1])
for i in range(noiseNum):
randX = random.randint(0, img.shape[0] - 1)
randY = random.randint(0, img.shape[1] - 1)
if random.randint(0, 1) == 0:
img[randX, randY] = 0
else:
img[randX, randY] = 255
return Image.fromarray(img)
# dataset for training
# The current loader is not using the normalized depth maps for training and test. If you use the normalized depth maps
# (e.g., 0 represents background and 1 represents foreground.), the performance will be further improved.
class SalObjDataset(data.Dataset):
def __init__(self, image_root, gt_root, depth_root, edge_root, trainsize):
self.trainsize = trainsize
self.images = [image_root + f for f in os.listdir(image_root) if f.endswith('.jpg')]
self.gts = [gt_root + f for f in os.listdir(gt_root) if f.endswith('.jpg')
or f.endswith('.png')]
self.depths = [depth_root + f for f in os.listdir(depth_root) if f.endswith('.bmp')
or f.endswith('.png') or f.endswith('.jpg')]
self.edges = [edge_root + f for f in os.listdir(edge_root) if f.endswith('.bmp')
or f.endswith('.png') or f.endswith('.jpg')]
self.images = sorted(self.images)
self.gts = sorted(self.gts)
self.depths = sorted(self.depths)
self.edges = sorted(self.edges)
self.filter_files()
self.size = len(self.images)
self.img_transform = transforms.Compose([
transforms.Resize((self.trainsize, self.trainsize)),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
self.gt_transform = transforms.Compose([
transforms.Resize((self.trainsize, self.trainsize)),
transforms.ToTensor()])
self.depths_transform = transforms.Compose(
[transforms.Resize((self.trainsize, self.trainsize)), transforms.ToTensor()])
self.edges_transform = transforms.Compose(
[transforms.Resize((self.trainsize, self.trainsize)), transforms.ToTensor()]
)
def __getitem__(self, index):
image = self.rgb_loader(self.images[index])
gt = self.binary_loader(self.gts[index])
depth = self.binary_loader(self.depths[index])
edge = self.binary_loader(self.edges[index])
image, gt, depth, edge = cv_random_flip(image, gt, depth, edge)
image, gt, depth, edge = randomCrop(image, gt, depth, edge)
image, gt, depth, edge = randomRotation(image, gt, depth, edge)
image = colorEnhance(image)
# gt=randomGaussian(gt)
gt = randomPeper(gt)
image = self.img_transform(image)
gt = self.gt_transform(gt)
depth = self.depths_transform(depth)
edge = self.edges_transform(edge)
return image, gt, depth, edge
def filter_files(self):
assert len(self.images) == len(self.gts) and len(self.gts) == len(self.images)
images = []
gts = []
depths = []
edges = []
for img_path, gt_path, depth_path, edge_path in zip(self.images, self.gts, self.depths, self.edges):
img = Image.open(img_path)
gt = Image.open(gt_path)
depth = Image.open(depth_path)
edge = Image.open(edge_path)
if img.size == gt.size and gt.size == depth.size and edge.size == img.size:
images.append(img_path)
gts.append(gt_path)
depths.append(depth_path)
edges.append(edge_path)
self.images = images
self.gts = gts
self.depths = depths
self.edges = edges
def rgb_loader(self, path):
with open(path, 'rb') as f:
img = Image.open(f)
return img.convert('RGB')
def binary_loader(self, path):
with open(path, 'rb') as f:
img = Image.open(f)
return img.convert('L')
def resize(self, img, gt, depth, edge):
assert img.size == gt.size and gt.size == depth.size and edge.size == img.size
w, h = img.size
if h < self.trainsize or w < self.trainsize:
h = max(h, self.trainsize)
w = max(w, self.trainsize)
return img.resize((w, h), Image.BILINEAR), gt.resize((w, h), Image.NEAREST), \
depth.resize((w, h),Image.NEAREST), edge.resize((w,h), Image.NEAREST)
else:
return img, gt, depth, edge
def __len__(self):
return self.size
# dataloader for training
def get_loader(image_root, gt_root, depth_root, edge_root, batchsize, trainsize, shuffle=True, num_workers=0, pin_memory=True):
dataset = SalObjDataset(image_root, gt_root, depth_root, edge_root, trainsize)
# print(image_root)
# print(gt_root)
# print(depth_root)
data_loader = data.DataLoader(dataset=dataset,
batch_size=batchsize,
shuffle=shuffle,
num_workers=num_workers,
pin_memory=pin_memory)
return data_loader
# test dataset and loader
class test_dataset:
def __init__(self, image_root, gt_root, depth_root, testsize):
self.testsize = testsize
self.images = [image_root + f for f in os.listdir(image_root) if f.endswith('.jpg')]
self.gts = [gt_root + f for f in os.listdir(gt_root) if f.endswith('.jpg')
or f.endswith('.png')]
self.depths = [depth_root + f for f in os.listdir(depth_root) if f.endswith('.bmp')
or f.endswith('.png')or f.endswith('.jpg')]
# self.edges = [edge_root + f for f in os.listdir(depth_root) if f.endswith('.bmp')
# or f.endswith('.png')or f.endswith('.jpg')]
self.images = sorted(self.images)
self.gts = sorted(self.gts)
self.depths = sorted(self.depths)
self.transform = transforms.Compose([
transforms.Resize((self.testsize, self.testsize)),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
self.gt_transform = transforms.ToTensor()
# self.gt_transform = transforms.Compose([
# transforms.Resize((self.trainsize, self.trainsize)),
# transforms.ToTensor()])
self.depths_transform = transforms.Compose(
[transforms.Resize((self.testsize, self.testsize)), transforms.ToTensor()])
# self.edges_transform = transforms.Compose(
# [transforms.Resize((self.testsize, self.testsize)), transforms.ToTensor()])
self.size = len(self.images)
self.index = 0
def load_data(self):
image = self.rgb_loader(self.images[self.index])
image = self.transform(image).unsqueeze(0)
gt = self.binary_loader(self.gts[self.index])
depth = self.binary_loader(self.depths[self.index])
depth = self.depths_transform(depth).unsqueeze(0)
name = self.images[self.index].split('/')[-1]
image_for_post = self.rgb_loader(self.images[self.index])
image_for_post = image_for_post.resize(gt.size)
if name.endswith('.jpg'):
name = name.split('.jpg')[0] + '.png'
self.index += 1
self.index = self.index % self.size
return image, gt, depth, name, np.array(image_for_post)
def rgb_loader(self, path):
with open(path, 'rb') as f:
img = Image.open(f)
return img.convert('RGB')
def binary_loader(self, path):
with open(path, 'rb') as f:
img = Image.open(f)
return img.convert('L')
def __len__(self):
return self.size