dataset_config.py
5.4 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
# Code for "TSM: Temporal Shift Module for Efficient Video Understanding"
# arXiv:1811.08383
# Ji Lin*, Chuang Gan, Song Han
# {jilin, songhan}@mit.edu, ganchuang@csail.mit.edu
import os
ROOT_DATASET = '/data1/action_1_images/' # '/data/jilin/'
def return_ucf101(modality):
filename_categories = 'labels/classInd.txt'
if modality == 'RGB':
root_data = ROOT_DATASET + 'images'
filename_imglist_train = 'file_list/ucf101_rgb_train_split_1.txt'
filename_imglist_val = 'file_list/ucf101_rgb_val_split_1.txt'
prefix = 'img_{:05d}.jpg'
elif modality == 'Flow':
root_data = ROOT_DATASET + 'UCF101/jpg'
filename_imglist_train = 'UCF101/file_list/ucf101_flow_train_split_1.txt'
filename_imglist_val = 'UCF101/file_list/ucf101_flow_val_split_1.txt'
prefix = 'flow_{}_{:05d}.jpg'
else:
raise NotImplementedError('no such modality:' + modality)
return filename_categories, filename_imglist_train, filename_imglist_val, root_data, prefix
def return_hmdb51(modality):
filename_categories = 51
if modality == 'RGB':
root_data = ROOT_DATASET + 'HMDB51/images'
filename_imglist_train = 'HMDB51/splits/hmdb51_rgb_train_split_1.txt'
filename_imglist_val = 'HMDB51/splits/hmdb51_rgb_val_split_1.txt'
prefix = 'img_{:05d}.jpg'
elif modality == 'Flow':
root_data = ROOT_DATASET + 'HMDB51/images'
filename_imglist_train = 'HMDB51/splits/hmdb51_flow_train_split_1.txt'
filename_imglist_val = 'HMDB51/splits/hmdb51_flow_val_split_1.txt'
prefix = 'flow_{}_{:05d}.jpg'
else:
raise NotImplementedError('no such modality:' + modality)
return filename_categories, filename_imglist_train, filename_imglist_val, root_data, prefix
def return_something(modality):
filename_categories = 'something/v1/category.txt'
if modality == 'RGB':
root_data = ROOT_DATASET + 'something/v1/20bn-something-something-v1'
filename_imglist_train = 'something/v1/train_videofolder.txt'
filename_imglist_val = 'something/v1/val_videofolder.txt'
prefix = '{:05d}.jpg'
elif modality == 'Flow':
root_data = ROOT_DATASET + 'something/v1/20bn-something-something-v1-flow'
filename_imglist_train = 'something/v1/train_videofolder_flow.txt'
filename_imglist_val = 'something/v1/val_videofolder_flow.txt'
prefix = '{:06d}-{}_{:05d}.jpg'
else:
print('no such modality:'+modality)
raise NotImplementedError
return filename_categories, filename_imglist_train, filename_imglist_val, root_data, prefix
def return_somethingv2(modality):
filename_categories = 'something/v2/category.txt'
if modality == 'RGB':
root_data = ROOT_DATASET + 'something/v2/20bn-something-something-v2-frames'
filename_imglist_train = 'something/v2/train_videofolder.txt'
filename_imglist_val = 'something/v2/val_videofolder.txt'
prefix = '{:06d}.jpg'
elif modality == 'Flow':
root_data = ROOT_DATASET + 'something/v2/20bn-something-something-v2-flow'
filename_imglist_train = 'something/v2/train_videofolder_flow.txt'
filename_imglist_val = 'something/v2/val_videofolder_flow.txt'
prefix = '{:06d}.jpg'
else:
raise NotImplementedError('no such modality:'+modality)
return filename_categories, filename_imglist_train, filename_imglist_val, root_data, prefix
def return_jester(modality):
filename_categories = 'jester/category.txt'
if modality == 'RGB':
prefix = '{:05d}.jpg'
root_data = ROOT_DATASET + 'jester/20bn-jester-v1'
filename_imglist_train = 'jester/train_videofolder.txt'
filename_imglist_val = 'jester/val_videofolder.txt'
else:
raise NotImplementedError('no such modality:'+modality)
return filename_categories, filename_imglist_train, filename_imglist_val, root_data, prefix
def return_kinetics(modality):
filename_categories = 400
if modality == 'RGB':
root_data = ROOT_DATASET + 'kinetics/images'
filename_imglist_train = 'kinetics/labels/train_videofolder.txt'
filename_imglist_val = 'kinetics/labels/val_videofolder.txt'
prefix = 'img_{:05d}.jpg'
else:
raise NotImplementedError('no such modality:' + modality)
return filename_categories, filename_imglist_train, filename_imglist_val, root_data, prefix
def return_dataset(dataset, modality):
dict_single = {'jester': return_jester, 'something': return_something, 'somethingv2': return_somethingv2,
'ucf101': return_ucf101, 'hmdb51': return_hmdb51,
'kinetics': return_kinetics}
if dataset in dict_single:
file_categories, file_imglist_train, file_imglist_val, root_data, prefix = dict_single[dataset](modality)
else:
raise ValueError('Unknown dataset '+dataset)
file_imglist_train = os.path.join(ROOT_DATASET, file_imglist_train)
file_imglist_val = os.path.join(ROOT_DATASET, file_imglist_val)
if isinstance(file_categories, str):
file_categories = os.path.join(ROOT_DATASET, file_categories)
with open(file_categories) as f:
lines = f.readlines()
categories = [item.rstrip() for item in lines]
else: # number of categories
categories = [None] * file_categories
n_class = len(categories)
print('{}: {} classes'.format(dataset, n_class))
return n_class, file_imglist_train, file_imglist_val, root_data, prefix