-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdataloaders.py
225 lines (198 loc) · 10.5 KB
/
dataloaders.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
"""Pytorch dataset objects that loads QMNIST-bags and Imagenette-bags datasets for experiments with/without within-bag sampling"""
import numpy as np
import torch
import torch.utils.data as data_utils
from torchvision import datasets, transforms
from pathlib import Path
import os
from PIL import Image
def list_files_in_folder(image_folder):
"""Lists file names in a given directory"""
list_of_files = []
for file in os.listdir(image_folder):
if os.path.isfile(os.path.join(image_folder, file)):
list_of_files.append(file)
return list_of_files
class QmnistBags(data_utils.Dataset):
def __init__(self, transform, sampling_size, train, valid, test, data_path, image_size, key_ins_digit):
self.train = train; self.valid = valid; self.test = test
self.transform = transform
self.sampling_size = sampling_size
self.image_size = image_size
self.key_ins_digit = key_ins_digit
if self.train==True:
self.datapath = os.path.join(data_path, 'train')
self.list_paths = self._list_bags_paths()
self.train_bags_list, self.train_labels_list, self.train_imgs_lists, self.bags_names_train = self._create_bags()
np.save(os.path.join(data_path, 'train_imgs_lists.npy'), np.asarray(self.train_imgs_lists))
elif self.valid==True:
self.datapath = os.path.join(data_path, 'valid')
self.list_paths = self._list_bags_paths()
self.valid_bags_list, self.valid_labels_list, self.valid_imgs_lists, self.bags_names_valid = self._create_bags()
np.save(os.path.join(data_path, 'valid_imgs_lists.npy'), np.asarray(self.valid_imgs_lists))
elif self.test==True:
self.datapath = os.path.join(data_path, 'test')
self.list_paths = self._list_bags_paths()
self.test_bags_list, self.test_labels_list, self.test_imgs_lists, self.bags_names_test = self._create_bags()
np.save(os.path.join(data_path,'test_imgs_lists.npy'), np.asarray(self.test_imgs_lists))
def _list_bags_paths(self):
list_paths = []
basePath = Path(self.datapath)
for child in basePath.iterdir():
if child.is_dir():
for grandchild in child.iterdir():
if grandchild.is_dir():
list_paths.append(grandchild)
return list_paths
def _create_bags(self):
all_bags = []
all_bags_labels = []
lists_imgs_all = []
bags_names_all=[]
for p in range(len(self.list_paths)):
list_img_names = list_files_in_folder(str(self.list_paths[p]))
basePath = Path(self.list_paths[p])
bags_names_all.append(np.asarray(int(basePath.parts[-1])))
list_img_names_bag = ()
for i in range(len(list_img_names)):
temp = (list_img_names[i],)
list_img_names_bag = list_img_names_bag+temp
lists_imgs_all.append(np.asarray(list_img_names_bag))
X = torch.empty(1, self.image_size[0], self.image_size[1], self.image_size[2])
y = torch.empty(1)
for i in range(len(list_img_names)):
img = Image.open(os.path.join(str(self.list_paths[p]), list_img_names[i]))
img_tensor = self.transform(img).float()
X = torch.cat((X,torch.unsqueeze(img_tensor, 0)),0)
if list_img_names[i][0]==self.key_ins_digit:
y = torch.cat((y,torch.ones(1)),0)
else:
y = torch.cat((y,torch.zeros(1)),0)
all_bags.append(X[1:,:,:,:])
all_bags_labels.append(y[1:])
return all_bags, all_bags_labels, lists_imgs_all, bags_names_all
def __len__(self):
return len(self.list_paths)
def __getitem__(self, index):
# Introduce sampling, with replacement for train, test and valid
max_number_of_imgs_in_bag = int(self.sampling_size)
if self.train:
bag = self.train_bags_list[index]
label = [max(self.train_labels_list[index]), self.train_labels_list[index]]
lists_of_names = self.train_imgs_lists[index]
bages_names = self.bags_names_train[index]
elif self.valid:
bag = self.valid_bags_list[index]
label = [max(self.valid_labels_list[index]), self.valid_labels_list[index]]
lists_of_names = self.valid_imgs_lists[index]
bages_names = self.bags_names_valid[index]
else:
bag = self.test_bags_list[index]
label = [max(self.test_labels_list[index]), self.test_labels_list[index]]
lists_of_names = self.test_imgs_lists[index]
bages_names = self.bags_names_test[index]
if bag.shape[0] > max_number_of_imgs_in_bag:
sample_indices = torch.randint(bag.shape[0], (max_number_of_imgs_in_bag,))
bag_sampled = bag[sample_indices,:,:,:]
label_sampled = [label[0],label[1][sample_indices]]
lists_of_names_sampled = [lists_of_names[x] for x in sample_indices]
bages_names_sampled = bages_names
else:
sample_indices = np.arange(bag.shape[0])
bag_sampled = bag
label_sampled = label
lists_of_names_sampled = lists_of_names
bages_names_sampled = bages_names
return bag_sampled, label_sampled, torch.as_tensor(sample_indices), torch.as_tensor(index), torch.as_tensor(bages_names_sampled)
class ImagenetteBags(data_utils.Dataset):
def __init__(self, transform, sampling_size, train, valid, test, image_size, data_path):
self.train = train; self.valid = valid; self.test = test
self.transform = transform
self.sampling_size = sampling_size
self.image_size = image_size
if self.train==True:
self.datapath = os.path.join(data_path, 'train')
self.list_paths = self._list_bags_paths()
self.train_bags_list, self.train_labels_list, self.train_imgs_lists, self.bags_names_train = self._create_bags()
print(self.train_bags_list[0].shape, self.train_labels_list[0].shape)
np.save(os.path.join(data_path, 'train_imgs_lists.npy'), np.asarray(self.train_imgs_lists))
elif self.valid==True:
self.datapath = os.path.join(data_path, 'valid')
self.list_paths = self._list_bags_paths()
self.valid_bags_list, self.valid_labels_list, self.valid_imgs_lists, self.bags_names_valid = self._create_bags()
np.save(os.path.join(data_path, 'valid_imgs_lists.npy'), np.asarray(self.valid_imgs_lists))
elif self.test==True:
self.datapath = os.path.join(data_path, 'test')
self.list_paths = self._list_bags_paths()
self.test_bags_list, self.test_labels_list, self.test_imgs_lists, self.bags_names_test = self._create_bags()
np.save(os.path.join(data_path,'test_imgs_lists.npy'), np.asarray(self.test_imgs_lists))
def _list_bags_paths(self):
list_paths = []
basePath = Path(self.datapath)
for child in basePath.iterdir():
if child.is_dir():
for grandchild in child.iterdir():
if grandchild.is_dir():
list_paths.append(grandchild)
return list_paths
def _create_bags(self):
all_bags = []
all_bags_labels = []
lists_imgs_all = []
bags_names_all=[]
for p in range(len(self.list_paths)):
list_img_names = list_files_in_folder(str(self.list_paths[p]))
basePath = Path(self.list_paths[p])
bags_names_all.append(np.asarray(int(basePath.parts[-1])))
print(basePath.parts[-3])
list_img_names_bag = ()
for i in range(len(list_img_names)):
temp = (list_img_names[i],)
list_img_names_bag = list_img_names_bag+temp
lists_imgs_all.append(np.asarray(list_img_names_bag))
X = torch.empty(1, self.image_size[0], self.image_size[1], self.image_size[2])
y = torch.empty(1)
for i in range(len(list_img_names)):
img = Image.open(os.path.join(str(self.list_paths[p]), list_img_names[i]))
img_tensor = self.transform(img).float()
X = torch.cat((X,torch.unsqueeze(img_tensor, 0)),0)
if list_img_names[i][0:6]=='keyins':
y = torch.cat((y,torch.ones(1)),0)
else:
y = torch.cat((y,torch.zeros(1)),0)
all_bags.append(X[1:,:,:,:])
all_bags_labels.append(y[1:])
return all_bags, all_bags_labels, lists_imgs_all, bags_names_all
def __len__(self):
return len(self.list_paths)
def __getitem__(self, index):
# Introduce sampling, with replacement for train, test and valid
max_number_of_imgs_in_bag = int(self.sampling_size)
if self.train:
bag = self.train_bags_list[index]
label = [max(self.train_labels_list[index]), self.train_labels_list[index]]
lists_of_names = self.train_imgs_lists[index]
bages_names = self.bags_names_train[index]
elif self.valid:
bag = self.valid_bags_list[index]
label = [max(self.valid_labels_list[index]), self.valid_labels_list[index]]
lists_of_names = self.valid_imgs_lists[index]
bages_names = self.bags_names_valid[index]
else:
bag = self.test_bags_list[index]
label = [max(self.test_labels_list[index]), self.test_labels_list[index]]
lists_of_names = self.test_imgs_lists[index]
bages_names = self.bags_names_test[index]
if bag.shape[0] > max_number_of_imgs_in_bag:
sample_indices = torch.randint(bag.shape[0], (max_number_of_imgs_in_bag,))
bag_sampled = bag[sample_indices,:,:,:]
label_sampled = [label[0],label[1][sample_indices]]
lists_of_names_sampled = [lists_of_names[x] for x in sample_indices]
bages_names_sampled = bages_names
else:
sample_indices = np.arange(bag.shape[0])
bag_sampled = bag
label_sampled = label
lists_of_names_sampled = lists_of_names
bages_names_sampled = bages_names
return bag_sampled, label_sampled, torch.as_tensor(sample_indices), torch.as_tensor(index), torch.as_tensor(bages_names_sampled)