-
Notifications
You must be signed in to change notification settings - Fork 27
/
preprocess_msrvttqa.py
178 lines (144 loc) · 6.55 KB
/
preprocess_msrvttqa.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
"""Preprocess the data of MSRVTT-QA."""
import os
import sys
import pandas as pd
from pandas import Series, DataFrame
import tables
import tensorflow as tf
from util.preprocess import VideoVGGExtractor
from util.preprocess import VideoC3DExtractor
from util.preprocess import prune_embedding
def extract_vgg(video_directory):
"""Extract VGG features."""
vgg_features = list()
# Session config.
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
sess_config.gpu_options.visible_device_list = '0'
with tf.Graph().as_default(), tf.Session(config=sess_config) as sess:
extractor = VideoVGGExtractor(20, sess)
for i in range(0, 10000):
video_path = os.path.join(
video_directory, 'video' + str(i) + '.mp4')
print('[VGG]', video_path)
vgg_features.append(extractor.extract(video_path))
# print(vgg_features[-1])
return vgg_features
def extract_c3d(video_directory):
"""Extract C3D features."""
c3d_features = list()
# Session config.
sess_config = tf.ConfigProto()
sess_config.gpu_options.allow_growth = True
sess_config.gpu_options.visible_device_list = '0'
with tf.Graph().as_default(), tf.Session(config=sess_config) as sess:
extractor = VideoC3DExtractor(20, sess)
for i in range(0, 10000):
video_path = os.path.join(
video_directory, 'video' + str(i) + '.mp4')
print('[C3D]', video_path)
c3d_features.append(extractor.extract(video_path))
# print(c3d_features[-1])
return c3d_features
def extract_video_feature(video_directory, feature_path):
"""Extract video features(vgg, c3d) and store in hdf5 file."""
h5file = tables.open_file(
feature_path, 'w', 'Extracted video features of the MSRVTT-QA dataset.')
vgg_features = extract_vgg(video_directory)
h5file.create_array('/', 'vgg', vgg_features, 'vgg16 feature')
c3d_features = extract_c3d(video_directory)
h5file.create_array('/', 'c3d', c3d_features, 'c3d feature')
h5file.close()
def create_answerset(trainqa_path, answerset_path):
"""Generate 1000 answer set from train_qa.json.
Args:
trainqa_path: path to train_qa.json.
answerset_path: generate answer set of mc_qa
"""
train_qa = pd.read_json(trainqa_path)
answer_freq = train_qa['answer'].value_counts()
answer_freq = DataFrame(answer_freq.iloc[0:1000])
answer_freq.to_csv(answerset_path, columns=[], header=False)
def create_vocab(trainqa_path, answerset_path, vocab_path):
"""Create the 8000 vocabulary based on questions in train split.
7999 most frequent words and 1 <UNK>.
Args:
trainqa_path: path to train_qa.json.
vocab_path: vocabulary file.
"""
vocab = dict()
train_qa = pd.read_json(trainqa_path)
# remove question whose answer is not in answerset
answerset = pd.read_csv(answerset_path, header=None)[0]
train_qa = train_qa[train_qa['answer'].isin(answerset)]
questions = train_qa['question'].values
for q in questions:
words = q.rstrip('?').split()
for word in words:
if len(word) >= 2:
vocab[word] = vocab.get(word, 0) + 1
vocab = Series(vocab)
vocab.sort_values(ascending=False, inplace=True)
vocab = DataFrame(vocab.iloc[0:7999])
vocab.loc['<UNK>'] = [0]
vocab.to_csv(vocab_path, columns=[], header=False)
def create_qa_encode(vttqa_path, vocab_path, answerset_path,
trainqa_encode_path, valqa_encode_path, testqa_encode_path):
"""Encode question/answer for generate batch faster.
In train split, remove answers not in answer set and convert question and answer
to one hot encoding. In val and test split, only convert question to one hot encoding.
"""
train_qa = pd.read_json(os.path.join(vttqa_path, 'train_qa.json'))
# remove question whose answer not in answer set
answerset = pd.read_csv(answerset_path, header=None)[0]
train_qa = train_qa[train_qa['answer'].isin(answerset)]
val_qa = pd.read_json(os.path.join(vttqa_path, 'val_qa.json'))
test_qa = pd.read_json(os.path.join(vttqa_path, 'test_qa.json'))
vocab = pd.read_csv(vocab_path, header=None)[0]
def _encode_question(row):
"""Map question to sequence of vocab id. 7999 for word not in vocab."""
question = row['question']
question_id = ''
words = question.rstrip('?').split()
for word in words:
if word in vocab.values:
question_id = question_id + \
str(vocab[vocab == word].index[0]) + ','
else:
question_id = question_id + '7999' + ','
return question_id.rstrip(',')
def _encode_answer(row):
"""Map answer to category id."""
answer = row['answer']
answer_id = answerset[answerset == answer].index[0]
return answer_id
print('start train split encoding.')
train_qa['question_encode'] = train_qa.apply(_encode_question, axis=1)
train_qa['answer_encode'] = train_qa.apply(_encode_answer, axis=1)
print('start val split encoding.')
val_qa['question_encode'] = val_qa.apply(_encode_question, axis=1)
print('start test split encoding.')
test_qa['question_encode'] = test_qa.apply(_encode_question, axis=1)
train_qa.to_json(trainqa_encode_path, 'records')
val_qa.to_json(valqa_encode_path, 'records')
test_qa.to_json(testqa_encode_path, 'records')
def main():
os.makedirs('data/msrvtt_qa')
extract_video_feature(os.path.join(sys.argv[1], 'video'),
'data/msrvtt_qa/video_feature_20.h5')
create_answerset(os.path.join(sys.argv[1], 'train_qa.json'),
'data/msrvtt_qa/answer_set.txt')
create_vocab(os.path.join(sys.argv[1], 'train_qa.json'),
'data/msrvtt_qa/answer_set.txt',
'data/msrvtt_qa/vocab.txt')
prune_embedding('data/msrvtt_qa/vocab.txt',
'util/glove.6B.300d.txt',
'data/msrvtt_qa/word_embedding.npy')
create_qa_encode(sys.argv[1],
'data/msrvtt_qa/vocab.txt',
'data/msrvtt_qa/answer_set.txt',
'data/msrvtt_qa/train_qa_encode.json',
'data/msrvtt_qa/val_qa_encode.json',
'data/msrvtt_qa/test_qa_encode.json')
if __name__ == '__main__':
main()