Repository navigation
Expand file tree
/
Copy pathsplit_data.py
More file actions
241 lines (199 loc) · 10.5 KB
/
Copy pathsplit_data.py
File metadata and controls
241 lines (199 loc) · 10.5 KB
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
'''
Splits BDD100k object detection train data into:
* IID: randomly split across num_clients (equal num of datapoints)
* non-IID: each client only gets data from one weather type
* val: only first 4 clients have val sets (each representing one weather type for non-IID)
'''
import os
import shutil
from distutils.dir_util import copy_tree
import random
def remove(path):
""" param <path> could either be relative or absolute. """
if os.path.isfile(path) or os.path.islink(path):
os.remove(path) # remove the file
print('Deleted previous file saved at {path}.')
elif os.path.isdir(path):
shutil.rmtree(path) # remove dir and all contains
print('Cleared old contents of folder {path}.')
else:
raise ValueError("file {} is not a file or dir.".format(path))
def create_folder(path):
if not os.path.exists(path):
os.makedirs(path)
else:
raise ValueError(f"{path} already exists.")
def move_num(src_path, dst_path, num):
files = os.listdir(src_path)
for file_name in random.sample(files, num):
shutil.move(os.path.join(src_path, file_name), dst_path)
def setup_data(dir_name = 'temp', num_labeled=1, num_unlabeled=3, data_split_type = 'ssfl', iid=True, val_split_names=['cloudy', 'overcast', 'rainy', 'snowy']):
'''
:param (string) data_split_type: one of ['centralized', 'ssl', 'ssfl']
:param (string list) val_split_names: default ['cloudy', 'overcast', 'rainy', 'snowy']
Remove old data folders
Create new data folders in following structure
data/
{dir_name}/
train/
ds_0_labeled/
labeled/
images/
trains/
labels/
trains/
(optional if labeled exists) unlabeled/
images/
trains/
ds_1_unlabeled/
ds_2_unlabeled/
.
.
val/
{cloudy}/
images/
valids/
labels/
valids/
{overcast}/
{rainy}/
{snowy}/
.
.
:return (dict) data_mapping: {{i}: {'train': <path>, 'val': <path>, 'target': <path>}}
'''
assert data_split_type in ['centralized', 'standalone', 'ssl', 'ssfl'], 'data_split_type must be one of [centralized, standalone, ssl, ssfl]'
assert num_labeled == 1, 'num_labeled != 1 not implemented'
data_dir = f'data/{dir_name}'
total_num_ds = 0
data_mapping = {}
# Remove and recreate
remove(data_dir)
create_folder(f'{data_dir}')
create_folder(f'{data_dir}/train')
# Create labeled train
if data_split_type == 'ssl' or data_split_type == 'centralized':
num_labeled = 1
num_unlabeled = 0
elif data_split_type == 'standalone':
num_labeled = len(val_split_names)
num_unlabeled = 0
for i in range(num_labeled):
create_folder(f'{data_dir}/train/ds_{i}_labeled')
create_folder(f'{data_dir}/train/ds_{i}_labeled/labeled')
create_folder(f'{data_dir}/train/ds_{i}_labeled/labeled/images')
create_folder(f'{data_dir}/train/ds_{i}_labeled/labeled/images/trains')
create_folder(f'{data_dir}/train/ds_{i}_labeled/labeled/labels')
create_folder(f'{data_dir}/train/ds_{i}_labeled/labeled/labels/trains')
if i == 0 and data_split_type == 'ssl':
# Also create unlabeled folder for first datasource
create_folder(f'{data_dir}/train/ds_{i}_unlabeled')
create_folder(f'{data_dir}/train/ds_{i}_unlabeled/unlabeled')
create_folder(f'{data_dir}/train/ds_{i}_unlabeled/unlabeled/images')
create_folder(f'{data_dir}/train/ds_{i}_unlabeled/unlabeled/images/trains')
data_mapping[i] = {
'train': f'{data_dir}/train/ds_{i}_labeled/labeled/images/trains',
'target': f'{data_dir}/train/ds_{i}_unlabeled/unlabeled/images/trains'
}
else:
data_mapping[i] = {
'train': f'{data_dir}/train/ds_{i}_labeled/labeled/images/trains',
'target': 'empty'
}
total_num_ds += 1
print(f'Created {total_num_ds} labeled train folder(s)')
# Create unlabeled train
for i in range(num_labeled, num_labeled+num_unlabeled):
create_folder(f'{data_dir}/train/ds_{i}_unlabeled')
create_folder(f'{data_dir}/train/ds_{i}_unlabeled/unlabeled')
create_folder(f'{data_dir}/train/ds_{i}_unlabeled/unlabeled/images')
create_folder(f'{data_dir}/train/ds_{i}_unlabeled/unlabeled/images/trains')
data_mapping[i] = {
'train': 'empty',
'target': f'{data_dir}/train/ds_{i}_unlabeled/unlabeled/images/trains'
}
total_num_ds += 1
print(f'Created {total_num_ds} unlabeled train folder(s)')
# Create val
create_folder(f'{data_dir}/val')
for i in range(len(val_split_names)):
create_folder(f'{data_dir}/val/{val_split_names[i]}')
create_folder(f'{data_dir}/val/{val_split_names[i]}/images')
create_folder(f'{data_dir}/val/{val_split_names[i]}/images/valids')
create_folder(f'{data_dir}/val/{val_split_names[i]}/labels')
create_folder(f'{data_dir}/val/{val_split_names[i]}/labels/valids')
# Val -- copy folder over
copy_tree(f'data/{val_split_names[i]}/images/valids', f'{data_dir}/val/{val_split_names[i]}/images/valids')
copy_tree(f'data/{val_split_names[i]}/labels/valids', f'{data_dir}/val/{val_split_names[i]}/labels/valids')
print(f'Created {len(val_split_names)} val folder(s)')
# Move train data to each folder and map val data
if data_split_type == 'centralized':
# Train -- move all to first datasource
copy_tree('data/cloudy/images/trains', f'{data_dir}/train/ds_0_labeled/labeled/images/trains')
copy_tree('data/cloudy/labels/trains', f'{data_dir}/train/ds_0_labeled/labeled/labels/trains')
copy_tree('data/overcast/images/trains', f'{data_dir}/train/ds_0_labeled/labeled/images/trains')
copy_tree('data/overcast/labels/trains', f'{data_dir}/train/ds_0_labeled/labeled/labels/trains')
copy_tree('data/rainy/images/trains', f'{data_dir}/train/ds_0_labeled/labeled/images/trains')
copy_tree('data/rainy/labels/trains', f'{data_dir}/train/ds_0_labeled/labeled/labels/trains')
copy_tree('data/snowy/images/trains', f'{data_dir}/train/ds_0_labeled/labeled/images/trains')
copy_tree('data/snowy/labels/trains', f'{data_dir}/train/ds_0_labeled/labeled/labels/trains')
# Val mapping -- one datasource for each val split
for i in range(len(val_split_names)):
if i in data_mapping:
data_mapping[i]['val'] = f'{data_dir}/val/{val_split_names[i]}/images/valids'
data_mapping[i]['test'] = f'{data_dir}/val/{val_split_names[i]}/images/valids'
else:
data_mapping[i] = {
'val': f'{data_dir}/val/{val_split_names[i]}/images/valids',
'test': f'{data_dir}/val/{val_split_names[i]}/images/valids',
'train': 'empty',
'target': 'empty'
}
elif data_split_type == 'ssl':
if iid:
# Train --
raise NotImplementedError
else:
# Train -- first datasource has all labeled and unlabeled in separate folders
copy_tree('data/cloudy/images/trains', f'{data_dir}/train/ds_0_labeled/labeled/images/trains')
copy_tree('data/cloudy/labels/trains', f'{data_dir}/train/ds_0_labeled/labeled/labels/trains')
# Train -- unlabeled
copy_tree('data/overcast/images/trains', f'{data_dir}/train/ds_0_unlabeled/unlabeled/images/trains')
copy_tree('data/rainy/images/trains', f'{data_dir}/train/ds_0_unlabeled/unlabeled/images/trains')
copy_tree('data/snowy/images/trains', f'{data_dir}/train/ds_0_unlabeled/unlabeled/images/trains')
# Val mapping -- one datasource for each val split
for i in range(len(val_split_names)):
if i in data_mapping:
data_mapping[i]['val'] = f'{data_dir}/val/{val_split_names[i]}/images/valids'
data_mapping[i]['test'] = f'{data_dir}/val/{val_split_names[i]}/images/valids'
else:
data_mapping[i] = {
'val': f'{data_dir}/val/{val_split_names[i]}/images/valids',
'test': f'{data_dir}/val/{val_split_names[i]}/images/valids',
'train': 'empty',
'target': 'empty'
}
elif data_split_type == 'ssfl':
if iid:
raise NotImplementedError
else:
# 1 labeled datasource
copy_tree('data/cloudy/images/trains', f'{data_dir}/train/ds_0_labeled/labeled/images/trains')
copy_tree('data/cloudy/labels/trains', f'{data_dir}/train/ds_0_labeled/labeled/labels/trains')
# Divide unlabeled datasource into n non-iid unlabeled datasources
assert num_unlabeled >= len(val_split_names), 'In non-IID SSFL, number of unlabeled must >= len(val_split_names)'
frac_per_ds = len(val_split_names) * 1.0 / num_unlabeled
num_to_move = {}
# First put all into len(val_split_names) unlabeled datasources
for val_i in range(len(val_split_names)):
copy_tree(f'data/{val_split_names[val_i]}/images/trains', f'{data_dir}/train/ds_{val_i}_unlabeled/unlabeled/images/trains')
temp_files = os.listdir(f'data/{val_split_names[val_i]}/images/trains')
num_to_move[f'{val_split_names[val_i]}'] = frac_per_ds * len(temp_files)
data_mapping[1]['val'] = f'{data_dir}/val/{val_split_names[val_i]}/images/valids'
data_mapping[1]['test'] = f'{data_dir}/val/{val_split_names[val_i]}/images/valids'
# Split across rest of unlabeled datasources
for unlabeled_i in range(len(val_split_names) + num_labeled, num_labeled + num_unlabeled):
def main():
setup_data()
if __name__ == '__main__':
main()