-
Notifications
You must be signed in to change notification settings - Fork 0
/
utils_datasets.py
251 lines (199 loc) · 9.83 KB
/
utils_datasets.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
# Copyright (c) Facebook, Inc. and its affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
#
import os
from pathlib import Path
from typing import Dict
import json
import numpy as np
import math
# change seed to pick up a different subset of random samples
seed = 99999
#seed = 12345
#seed = 19
def get_subset_data(video_names,annotations,num_of_examples):
examples_per_class = int(math.ceil(num_of_examples / len(set(annotations))))
print("original data length", len(video_names) )
print("subset data", examples_per_class,num_of_examples)
random_state = np.random.RandomState(seed)
annotations = np.array(annotations)
video_names = np.array(video_names )
subset_video_names = []
subset_annotations = []
for class_label in set(annotations): # sample uniformly from each class
subset_indexes = np.where(annotations == class_label)
#print("shape",subset_indexes[0].shape[0])
if examples_per_class > subset_indexes[0].shape[0]:
ran_indicies = np.array(random_state.choice(subset_indexes[0].shape[0],subset_indexes[0].shape[0],replace=False))
else:
ran_indicies = np.array(random_state.choice(subset_indexes[0].shape[0],examples_per_class,replace=False))
indicies_100 = (subset_indexes[0][ran_indicies])
temp_annotations =annotations[indicies_100]
temp_names= video_names[indicies_100]
subset_video_names.extend(temp_names)
subset_annotations.extend(temp_annotations)
video_names = list(subset_video_names)
annotations = list(subset_annotations)
print(len(video_names),len(annotations))
#print(video_names)
#print(annotations)
return video_names, annotations
def get_filenames_and_labels_ucf(data_root,subset,num_of_examples=0):
DATA_PATH = data_root + 'UCF-101/'
ANNO_PATH = data_root + 'ucfTrainTestlist'
filenames = []
labels = []
classes_fn = f'{ANNO_PATH}/classInd.txt'
classes = [l.strip().split()[1] for l in open(classes_fn)]
if 'train' in subset:
filenames = [ln.strip().split()[0] for ln in open(f'{ANNO_PATH}/trainlist01.txt')]
labels = [fn.split('/')[0] for fn in filenames]
labels = [classes.index(cls) for cls in labels]
filenames = [DATA_PATH+ name for name in filenames ]
else:
filenames = [ln.strip().split()[0] for ln in open(f'{ANNO_PATH}/testlist01.txt')]
labels = [fn.split('/')[0] for fn in filenames]
labels = [classes.index(cls) for cls in labels]
filenames = [DATA_PATH+ name for name in filenames ]
if 'train' in subset and num_of_examples!=0:
filenames, labels = get_subset_data(filenames,labels,num_of_examples)
return filenames,labels
def get_filenames_and_labels_gym(data_root,subset,num_of_examples=0):
DATA_PATH = data_root + 'subactions/'
ANNO_PATH = data_root + 'annotations'
filenames = []
labels = []
if 'train' in subset:
for ln in open(f'{ANNO_PATH}/gym99_train.txt'):
file_name, label = ln.strip().split()[0][0:-3]+'avi',int(ln.strip().split()[1])
if os.path.isfile(DATA_PATH+'/'+file_name):
#print(file_name,label)
filenames.append(DATA_PATH+'/'+file_name)
labels.append(label)
else:
for ln in open(f'{ANNO_PATH}/gym99_val.txt'):
file_name, label = ln.strip().split()[0][0:-3]+'avi',int(ln.strip().split()[1])
if os.path.isfile(DATA_PATH+'/'+file_name):
filenames.append(DATA_PATH+'/'+file_name)
labels.append(label)
if 'train' in subset and num_of_examples!=0:
filenames, labels = get_subset_data(filenames,labels,num_of_examples)
return filenames,labels
def get_filenames_and_labels_gym_ub_s1(data_root,subset):
DATA_PATH = data_root + 'subactions/'
ANNO_PATH = data_root + 'annotations'
filenames = []
labels = []
action_classes_to_include = list(range(74,89)) # set UB-S1
if 'train' in subset:
for ln in open(f'{ANNO_PATH}/gym99_train.txt'):
file_name, label = ln.strip().split()[0][0:-3]+'avi',int(ln.strip().split()[1])
if os.path.isfile(DATA_PATH+'/'+file_name):
if label in action_classes_to_include:
filenames.append(DATA_PATH+'/'+file_name)
label = label - 74 # off set labels to start from 0
labels.append(label)
else:
for ln in open(f'{ANNO_PATH}/gym99_val.txt'):
file_name, label = ln.strip().split()[0][0:-3]+'avi',int(ln.strip().split()[1])
if os.path.isfile(DATA_PATH+'/'+file_name):
if label in action_classes_to_include:
filenames.append(DATA_PATH+'/'+file_name)
label = label - 74 # off set labels to start from 0
labels.append(label)
return filenames,labels
def get_filenames_and_labels_gym_fx_s1(data_root,subset):
DATA_PATH = data_root + 'subactions/'
ANNO_PATH = data_root + 'annotations'
filenames = []
labels = []
action_classes_to_include = [6,7,8,9,10,11,12,13,14,15,16] # set FX-S1
if 'train' in subset:
for ln in open(f'{ANNO_PATH}/gym99_train.txt'):
file_name, label = ln.strip().split()[0][0:-3]+'avi',int(ln.strip().split()[1])
if os.path.isfile(DATA_PATH+'/'+file_name):
#print(file_name,label)
if label in action_classes_to_include:
filenames.append(DATA_PATH+'/'+file_name)
label = label - 6
labels.append(label)
else:
for ln in open(f'{ANNO_PATH}/gym99_val.txt'):
file_name, label = ln.strip().split()[0][0:-3]+'avi',int(ln.strip().split()[1])
if os.path.isfile(DATA_PATH+'/'+file_name):
if label in action_classes_to_include:
filenames.append(DATA_PATH+'/'+file_name)
label = label - 6
labels.append(label)
return filenames,labels
def get_filenames_and_labels_diving(data_root,subset):
DATA_PATH = data_root + 'videos'
ANNO_PATH = data_root + 'labels'
filenames = []
labels = []
if 'train' in subset:
video_list_path = f'{ANNO_PATH}/Diving48_V2_train.json'
#with open(video_list_path) as f:
with open(video_list_path,encoding='utf-8') as f:
video_infos = json.load(f)
for video_info in video_infos:
video = video_info['vid_name']
video_name = f'{video}.avi'
label = int(video_info['label'])
if os.path.isfile(DATA_PATH+'/'+video_name):
filenames.append(DATA_PATH+'/'+video_name)
labels.append(label)
else:
video_list_path = f'{ANNO_PATH}/Diving48_V2_test.json'
#with open(video_list_path) as f:
with open(video_list_path,encoding='utf-8') as f:
video_infos = json.load(f)
for video_info in video_infos:
video = video_info['vid_name']
video_name = f'{video}.avi'
label = int(video_info['label'])
if os.path.isfile(DATA_PATH+'/'+video_name):
filenames.append(DATA_PATH+'/'+video_name)
labels.append(label)
return filenames,labels
def read_class_idx(annotation_dir: Path) -> Dict[str, str]:
class_ind_path = annotation_dir+'/something-something-v2-labels.json'
with open(class_ind_path) as f:
class_dict = json.load(f)
return class_dict
def get_filenames_and_labels_ssv2(data_root,subset):
DATA_PATH = data_root + 'something-something-v2-videos_avi'
ANNO_PATH = data_root + 'something-something-v2-annotations/'
class_idx_dict = read_class_idx(ANNO_PATH)
filenames = []
labels = []
if 'train' in subset:
video_list_path = f'{ANNO_PATH}/something-something-v2-train.json'
#with open(video_list_path) as f:
with open(video_list_path,encoding='utf-8') as f:
video_infos = json.load(f)
for video_info in video_infos:
video = int(video_info['id'])
video_name = f'{video}.avi'
class_name = video_info['template'].replace('[', '').replace(']', '')
class_index = int(class_idx_dict[class_name])
if os.path.isfile(DATA_PATH+'/'+video_name):
filenames.append(DATA_PATH+'/'+video_name)
labels.append(class_index)
else:
video_list_path = f'{ANNO_PATH}/something-something-v2-validation.json'
#with open(video_list_path) as f:
with open(video_list_path,encoding='utf-8') as f:
video_infos = json.load(f)
for video_info in video_infos:
video = int(video_info['id'])
video_name = f'{video}.avi'
class_name = video_info['template'].replace('[', '').replace(']', '')
class_index = int(class_idx_dict[class_name])
if os.path.isfile(DATA_PATH+'/'+video_name):
filenames.append(DATA_PATH+'/'+video_name)
labels.append(class_index)
return filenames,labels