-
Notifications
You must be signed in to change notification settings - Fork 1
/
extract_triplet.py
199 lines (168 loc) · 9.16 KB
/
extract_triplet.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
from datetime import datetime
from dataloader.outdoor_data import ActionsDataLoader
from models.multimodal import FuseDecoder
from models.unet_sound2 import UNetSound
from models.unet_architecture_energy import UNetE
from models.unet_noconc import UNetAc
from models.unet_architecture_noconc import UNet
import numpy as np
import tensorflow as tf
import os
import cv2
import matplotlib.pyplot as plt
flags = tf.app.flags
slim = tf.contrib.slim
flags.DEFINE_string('model', None, 'Model type, it can AudioCoeff')
flags.DEFINE_integer('temporal_pooling', 0, 'Temporal pooling')
flags.DEFINE_string('train_file', None, 'File for training data')
flags.DEFINE_string('init_checkpoint', None, 'Checkpoint file for model initialization')
flags.DEFINE_integer('num_classes', 9, 'Number of classes')
flags.DEFINE_integer('batch_size', 2, 'Batch size choose')
flags.DEFINE_integer('nr_frames', 1, 'Number of frames') # 12*FLAGS.sample_length max
flags.DEFINE_integer('sample_length', 1, 'Length in seconds of a sequence sample')
flags.DEFINE_integer('probability', 1, 'Use vae')
flags.DEFINE_string('datatype', 'outdoor', 'music or outdoor or old')
FLAGS = flags.FLAGS
'''Extract features old'''
def main(_):
plotdecodeimages()
def plotdecodeimages():
dataset = FLAGS.train_file.split('/')[-1]
dataset = dataset.split('.')[0]
s = FLAGS.init_checkpoint.split('/')[-1]
name = (s.split('_')[1]).split('.ckpt')[0]
nameac = '{}_{}_{}'.format(dataset, 'Ac', name)
nameaudio = '{}_{}_{}'.format(dataset, 'Audio', name)
nameimages = '{}_{}_{}'.format(dataset, 'Video', name)
data_dirac = str.join('/', FLAGS.init_checkpoint.split('/')[:-1] + [nameac])
data_diraudio = str.join('/', FLAGS.init_checkpoint.split('/')[:-1] + [nameaudio])
data_dirimages = str.join('/', FLAGS.init_checkpoint.split('/')[:-1] + [nameimages])
num_classes = FLAGS.num_classes
temporal_pooling = FLAGS.temporal_pooling
nr_frames = FLAGS.nr_frames
random_pick = True
build_spectrogram = (FLAGS.model == 'AudioCoefficients' or FLAGS.model == 'ResNet50' or FLAGS.model == 'HearNet'
or FLAGS.model == 'UNet' or FLAGS.model == 'ResNet18_v1')
normalize = FLAGS.model == 'HearNet'
# Create data loaders according to the received program arguments
print('{} - Creating data loaders'.format(datetime.now()))
modalities = []
modalities.append(0)
modalities.append(1)
modalities.append(2)
with tf.device('/cpu:0'):
train_data = ActionsDataLoader(FLAGS.train_file, 'inference', batch_size=FLAGS.batch_size, num_epochs=1, sample_length=1,
datakind='outdoor', buffer_size=10, shuffle=False,
normalize=normalize, build_spectrogram=build_spectrogram, correspondence=0,
random_pick=random_pick, modalities=modalities, nr_frames=FLAGS.nr_frames)
data_size = train_data.num_samples
# Build model
print('{} - Building model'.format(datetime.now()))
with tf.device('/gpu:0'):
modelimages = UNet(input_shape=[224, 298, 3])
modelaudio = UNetSound(input_shape=[99, 257, 1])
modelac = UNetAc(input_shape=[36, 48, 12])
# model = UNetE(input_shape=[36, 48, 1])
handle = tf.placeholder(tf.string, shape=())
iterator = tf.data.Iterator.from_string_handle(handle, train_data.data.output_types,
train_data.data.output_shapes)
train_iterat = train_data.data.make_initializable_iterator()
next_batch = iterator.get_next()
logenergy = tf.slice(next_batch[0], [0, 0, 0, 0, 0], [-1, 1, 36, 48, 1])
logenergy = tf.reshape(logenergy, shape=[-1, 36, 48, 1])
mfcc = tf.reshape(next_batch[1], shape=[-1, 99, 257, 1])
images = tf.reshape(next_batch[2], shape=[-1, 224, 298, 3])
acoustic = tf.reshape(next_batch[0], shape=[-1, 36, 48, 12])
logenergy = logenergy - tf.reduce_min(logenergy, axis=[1, 2], keep_dims=True)
logenergy = logenergy / tf.reduce_max(logenergy, axis=[1, 2], keep_dims=True)
# mfcc = mfcc - tf.reduce_min(mfcc, axis=[1, 2], keep_dims=True)
# mfcc = mfcc / tf.reduce_max(mfcc, axis=[1, 2], keep_dims=True)
if FLAGS.datatype == 'music':
num_actions = 9
num_locations = 11 # maximum number of videos for a class
else: # self.datakind == 'outdoor':
num_actions = 10
num_locations = 61
num_embedding = 128
labels = tf.reshape(next_batch[3], shape=[-1, num_actions])
scenario = tf.reshape(next_batch[4], shape=[-1, num_locations])
modelac._build_model(acoustic)
modelaudio._build_model(mfcc)
modelimages._build_model(images)
# samples = tf.random_normal([tf.shape(model.variance)[0], tf.shape(model.variance)[1]], 0, 1,
# dtype=tf.float32)
# guessed_z = model.mean + (model.variance * samples)
extractedac = modelac.network['features']
extractedaudio = modelaudio.network['features']
extractedvideo = modelimages.network['features']
#FLAGS.model == 'UNet'
var_listac = slim.get_variables(modelac.scope + '/')
var_listaudio = slim.get_variables(modelaudio.scope + '/')
var_listimages = slim.get_variables(modelimages.scope + '/')
if os.path.exists(data_dirac):
print("Features already computed!")
else:
os.makedirs(data_dirac) # mkdir creates one directory, makedirs all intermediate directories
if os.path.exists(data_diraudio):
print("Features already computed!")
else:
os.makedirs(data_diraudio) # mkdir creates one directory, makedirs all intermediate directories
if os.path.exists(data_dirimages):
print("Features already computed!")
else:
os.makedirs(data_dirimages) # mkdir creates one directory, makedirs all intermediate directories
total_size = 0
batch_count = 0
dataset_list_featuresac = np.zeros([data_size, 9, 12, num_embedding], dtype=float)
dataset_list_featuresaudio = np.zeros([data_size, 6, 16, num_embedding], dtype=float)
dataset_list_featuresimages = np.zeros([data_size, 14, 18, num_embedding], dtype=float)
dataset_labels = np.zeros([data_size, num_actions], dtype=int)
dataset_scenario = np.zeros([data_size, num_locations], dtype=int)
print('{} - Starting'.format(datetime.now()))
with tf.Session(
config=tf.ConfigProto(allow_soft_placement=True, gpu_options=tf.GPUOptions(allow_growth=True))) as session:
train_handle = session.run(train_iterat.string_handle())
saver = tf.train.Saver(var_list=var_listac + var_listaudio + var_listimages)
saver.restore(session, FLAGS.init_checkpoint)
print('{} - Done'.format(datetime.now()))
#variables_in_checkpoint = tf.train.list_variables('path.ckpt')
session.run(train_iterat.initializer)
while True:
try:
labels_data, scenario_data, featuresac, featuresaudio, featuresimages = session.run(
[labels, scenario, extractedac, extractedaudio, extractedvideo],
feed_dict={handle: train_handle,
modelac.network['keep_prob']: 1.0,
modelac.network['is_training']: 0,
modelaudio.network['keep_prob']: 1.0,
modelaudio.network['is_training']: 0,
modelimages.network['keep_prob']: 1.0,
modelimages.network['is_training']: 0})
batchnum = labels_data.shape[0]
# copy block of data
dataset_list_featuresimages[total_size:total_size + batchnum, :] = featuresimages
dataset_list_featuresaudio[total_size:total_size + batchnum, :] = featuresaudio
dataset_list_featuresac[total_size:total_size + batchnum, :] = featuresac
dataset_labels[total_size:total_size + batchnum, :] = labels_data
dataset_scenario[total_size:total_size + batchnum, :] = scenario_data
# increase number of data
total_size += batchnum
end_time = datetime.now()
print('{} samples'.format(total_size))
except tf.errors.OutOfRangeError:
break
batch_count += 1
print('{}'.format(data_size))
print('{} - Completed, got {} samples'.format(datetime.now(), total_size))
np.save('{}/{}_data.npy'.format(data_dirac, dataset), dataset_list_featuresac)
np.save('{}/{}_labels.npy'.format(data_dirac, dataset), dataset_labels)
np.save('{}/{}_scenario.npy'.format(data_dirac, dataset), dataset_scenario)
np.save('{}/{}_data.npy'.format(data_diraudio, dataset), dataset_list_featuresaudio)
np.save('{}/{}_labels.npy'.format(data_diraudio, dataset), dataset_labels)
np.save('{}/{}_scenario.npy'.format(data_diraudio, dataset), dataset_scenario)
np.save('{}/{}_data.npy'.format(data_dirimages, dataset), dataset_list_featuresimages)
np.save('{}/{}_labels.npy'.format(data_dirimages, dataset), dataset_labels)
np.save('{}/{}_scenario.npy'.format(data_dirimages, dataset), dataset_scenario)
if __name__ == '__main__':
flags.mark_flags_as_required(['train_file'])
tf.app.run()