-
Notifications
You must be signed in to change notification settings - Fork 6
/
dataset.py
275 lines (234 loc) · 10.5 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
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
265
266
267
268
269
270
271
272
273
274
import os
from PIL import Image
import torch
from torch.utils import data
import transforms as trans
import random
from parameter import *
class CoData_Train(data.Dataset):
def __init__(self, img_root, gt_root, img_root_coco, gt_root_coco, img_size, transform, t_transform, label_32_transform, label_64_transform, label_128_transform, max_num):
# Path Pool
self.img_root = img_root # root
self.gt_root = gt_root
self.dirs = os.listdir(img_root) # all dir
self.img_dir_paths = list( # [img_root+dir1, ..., img_root+dir2]
map(lambda x: os.path.join(img_root, x), self.dirs))
self.gt_dir_paths = list( # [gt_root+dir1, ..., gt_root+dir2]
map(lambda x: os.path.join(gt_root, x), self.dirs))
self.img_name_list = [os.listdir(idir) for idir in self.img_dir_paths
] # [[name00,..., 0N],..., [M0,..., MN]]
self.gt_name_list = [
map(lambda x: x[:-3] + 'png', iname_list)
for iname_list in self.img_name_list
]
self.img_path_list = [
list(
map(lambda x: os.path.join(self.img_dir_paths[idx], x),
self.img_name_list[idx]))
for idx in range(len(self.img_dir_paths))
] # [[impath00,..., 0N],..., [M0,..., MN]]
self.gt_path_list = [
list(
map(lambda x: os.path.join(self.gt_dir_paths[idx], x),
self.gt_name_list[idx]))
for idx in range(len(self.gt_dir_paths))
] # [[gtpath00,..., 0N],..., [M0,..., MN]]
self.nclass = len(self.dirs)
# CoCo Path Pool
self.img_root_coco = img_root_coco # root
self.gt_root_coco = gt_root_coco
self.dirs_coco = os.listdir(img_root_coco) # all dir
self.img_dir_paths_coco = list( # [img_root+dir1, ..., img_root+dir2]
map(lambda x: os.path.join(img_root_coco, x), self.dirs_coco))
self.gt_dir_paths_coco = list( # [gt_root+dir1, ..., gt_root+dir2]
map(lambda x: os.path.join(gt_root_coco, x), self.dirs_coco))
self.img_name_list_coco = [os.listdir(idir) for idir in self.img_dir_paths_coco
] # [[name00,..., 0N],..., [M0,..., MN]]
self.gt_name_list_coco = [
map(lambda x: x[:-3] + 'png', iname_list)
for iname_list in self.img_name_list_coco
]
self.img_path_list_coco = [
list(
map(lambda x: os.path.join(self.img_dir_paths_coco[idx], x),
self.img_name_list_coco[idx]))
for idx in range(len(self.img_dir_paths_coco))
] # [[impath00,..., 0N],..., [M0,..., MN]]
self.gt_path_list_coco = [
list(
map(lambda x: os.path.join(self.gt_dir_paths_coco[idx], x),
self.gt_name_list_coco[idx]))
for idx in range(len(self.gt_dir_paths_coco))
] # [[gtpath00,..., 0N],..., [M0,..., MN]]
self.nclass_coco = len(self.dirs_coco)
# Other Hyperparameters
self.size = img_size
self.cat_size = int(img_size * 2)
self.sizes = [img_size, img_size]
self.transform = transform
self.t_transform = t_transform
self.label_32_transform = label_32_transform
self.label_64_transform = label_64_transform
self.label_128_transform = label_128_transform
self.max_num = max_num
def __getitem__(self, item):
if random.random() < 0.5:
# select coco data
flag = False
sel = item%(self.nclass_coco-1)
img_paths = self.img_path_list_coco[sel]
gt_paths = self.gt_path_list_coco[sel]
else:
# select from our dataset
flag = True
img_paths = self.img_path_list[item]
gt_paths = self.gt_path_list[item]
num = len(img_paths)
if num > self.max_num:
sampled_list = random.sample(range(num), self.max_num)
new_img_paths = [img_paths[i] for i in sampled_list]
img_paths = new_img_paths
new_gt_paths = [gt_paths[i] for i in sampled_list]
gt_paths = new_gt_paths
num = self.max_num
imgs = torch.Tensor(num, 3, self.sizes[0], self.sizes[1])
gts = torch.Tensor(num, 1, self.sizes[0], self.sizes[1])
gts_32 = torch.Tensor(num, 1, img_size//8, img_size//8)
gts_64 = torch.Tensor(num, 1, img_size//4, img_size//4)
gts_128 = torch.Tensor(num, 1, img_size//2, img_size//2)
subpaths = []
ori_sizes = []
for idx in range(num):
if flag:
# data from our dataset
# random replace to syn img or do not replace
select_num = random.randint(1, 5)
if select_num == 4:
# select original img
img_path = img_paths[idx]
gt_path = gt_paths[idx]
if 1 <= select_num <= 3:
# select syn img
img_path = img_paths[idx].split('.jpg')[0] + '_syn' + str(select_num) + '.png'
img_path = img_syn_root + img_path.split('/img/')[1]
gt_path = gt_paths[idx]
if select_num == 5:
# select reverse syn img
select_reverse_num = random.randint(1, 3)
img_path = img_paths[idx].split('.jpg')[0] + '_ReverseSyn' + str(select_reverse_num) + '.png'
tmp = img_path.split('/img/')[1]
img_path = img_ReverseSyn_root + tmp
gt_path = gt_ReverseSyn_root + tmp
else:
# data from coco
img_path = img_paths[idx]
gt_path = gt_paths[idx]
img = Image.open(img_path).convert('RGB')
gt = Image.open(gt_path).convert('L')
subpaths.append(
os.path.join(img_paths[idx].split('/')[-2],
img_paths[idx].split('/')[-1][:-4] + '.png'))
ori_sizes.append((img.size[1], img.size[0]))
random_size = scale_size
new_img = trans.Scale((random_size, random_size))(img)
new_gt = trans.Scale((random_size, random_size), interpolation=Image.NEAREST)(gt)
# random crop
w, h = new_img.size
if w != img_size and h != img_size:
x1 = random.randint(0, w - img_size)
y1 = random.randint(0, h - img_size)
new_img = new_img.crop((x1, y1, x1 + img_size, y1 + img_size))
new_gt = new_gt.crop((x1, y1, x1 + img_size, y1 + img_size))
# random flip
if random.random() < 0.5:
new_img = new_img.transpose(Image.FLIP_LEFT_RIGHT)
new_gt = new_gt.transpose(Image.FLIP_LEFT_RIGHT)
new_img = self.transform(new_img)
gt_256 = self.t_transform(new_gt)
gt_32 = self.label_32_transform(new_gt)
gt_64 = self.label_64_transform(new_gt)
gt_128 = self.label_128_transform(new_gt)
imgs[idx] = new_img
gts[idx] = gt_256
gts_128[idx] = gt_128
gts_64[idx] = gt_64
gts_32[idx] = gt_32
return imgs, gts, gts_128, gts_64, gts_32, subpaths, ori_sizes
def __len__(self):
return len(self.dirs)
class CoData_Test(data.Dataset):
def __init__(self, img_root, img_size, transform):
class_list = os.listdir(img_root)
self.sizes = [img_size, img_size]
self.transform = transform
self.img_dirs = list(
map(lambda x: os.path.join(img_root, x), class_list))
def __getitem__(self, item):
names = os.listdir(self.img_dirs[item])
num = len(names)
img_paths = list(
map(lambda x: os.path.join(self.img_dirs[item], x), names))
imgs = torch.Tensor(num, 3, self.sizes[0], self.sizes[1])
subpaths = []
ori_sizes = []
for idx in range(num):
img = Image.open(img_paths[idx]).convert('RGB')
subpaths.append(
os.path.join(img_paths[idx].split('/')[-2],
img_paths[idx].split('/')[-1][:-4] + '.png'))
ori_sizes.append((img.size[1], img.size[0]))
img = self.transform(img)
imgs[idx] = img
return imgs, subpaths, ori_sizes
def __len__(self):
return len(self.img_dirs)
def get_loader(img_root, img_size, batch_size, gt_root=None, max_num=float('inf'), mode='train', num_thread=1, pin=False):
shuffle = False
mean = torch.Tensor(3, img_size, img_size)
mean[0, :, :] = 122.675 # R
mean[1, :, :] = 116.669 # G
mean[2, :, :] = 104.008 # B
mean_bgr = torch.Tensor(3, img_size, img_size)
mean_bgr[0, :, :] = 104.008 # B
mean_bgr[1, :, :] = 116.669 # G
mean_bgr[2, :, :] = 122.675 # R
if mode == 'train':
transform = trans.Compose([
# trans.ToTensor image -> [0,255]
trans.ToTensor_BGR(),
trans.Lambda(lambda x: x - mean_bgr)
])
t_transform = trans.Compose([
# transform.ToTensor label -> [0,1]
trans.ToTensor(),
])
label_32_transform = trans.Compose([
trans.Scale((img_size//8, img_size//8), interpolation=Image.NEAREST),
trans.ToTensor(),
])
label_64_transform = trans.Compose([
trans.Scale((img_size//4, img_size//4), interpolation=Image.NEAREST),
trans.ToTensor(),
])
label_128_transform = trans.Compose([
trans.Scale((img_size//2, img_size//2), interpolation=Image.NEAREST),
trans.ToTensor(),
])
shuffle = True
else:
transform = trans.Compose([
trans.Scale((img_size, img_size)),
trans.ToTensor_BGR(),
trans.Lambda(lambda x: x - mean_bgr)
])
t_transform = trans.Compose([
trans.Scale((img_size, img_size), interpolation=Image.NEAREST),
trans.ToTensor(),
])
if mode == 'train':
dataset = CoData_Train(img_root, gt_root, img_root_coco, gt_root_coco, img_size, transform, t_transform,
label_32_transform, label_64_transform, label_128_transform, max_num)
else:
dataset = CoData_Test(img_root, img_size, transform)
data_loader = data.DataLoader(dataset=dataset, batch_size=batch_size, shuffle=shuffle, num_workers=num_thread, pin_memory=pin)
return data_loader