def model_graph_asym(x, y_):
with tf.name_scope('fc_1'):
#Add the third densely connected layer
w_fc1=weight_variable([x.get_shape().as_list()[1],2000])
b_fc1=bias_variable([2000])
h_fc1=tflearn.relu(tf.matmul(x,w_fc1)+b_fc1)
with tf.name_scope('fc_2'):
#Add the third densely connected layer
w_fc2=weight_variable([
h_fc1.get_shape().as_list()[1],2000])
b_fc2=bias_variable([2000])
h_fc1=tflearn.relu(tf.matmul(h_fc1,w_fc2)+b_fc2)
with tf.name_scope('softmax_layer'):
w_s1=weight_variable([1500,1])
w_s2=weight_variable([502,1])
b_s=bias_variable([y_.get_shape().as_list()[1]])
y_conv_logit1 = tf.matmul(h_fc1[:,:1500], w_s1)
y_conv_logit2 = tf.matmul(
tf.concat([h_fc1[:,1500:], x], 1), w_s2)
y_conv_logit=tf.concat([y_conv_logit1,
y_conv_logit2],1)+b_s
y_conv=tf.nn.softmax(y_conv_logit)
return y_conv_logit, y_conv
def AuxiliaryHeadCIFAR(x,class_num): x=tflearn.relu(x) x=slim.avg_pool2d(x,[5,5], stride=3,padding='SAME') x=slim.conv2d(x,128,[1,1]) x=slim.batch_norm(x) x=tflearn.relu(x) x=slim.conv2d(x,768,[2,2]) x=slim.batch_norm(x) x=tflearn.relu(x) x=slim.flatten(x) x=slim.fully_connected(x,class_num, activation_fn=None) return x
def FactorizedReduce(x, c_out):
x = tflearn.relu(x)
conv1 = slim.conv2d(x, c_out // 2, [1, 1], stride=[2, 2])
conv2 = slim.conv2d(x[:, 1:, 1:, :], c_out // 2, [1, 1], stride=[2, 2])
x = tf.concat([conv1, conv2], -1)
x = slim.batch_norm(x)
return x
def DilConv(x, C_out, kernel_size, stride, rate):
x = tflearn.relu(x)
x = slim.separable_convolution2d(x,
C_out,
kernel_size,
depth_multiplier=1,
stride=stride)
x = slim.batch_norm(x)
return x
def model_test(x, is_training, class_num=10):
with tf.variable_scope("lw", reuse=tf.AUTO_REUSE):
with slim.arg_scope([slim.conv2d],
normalizer_fn=None,
activation_fn=tflearn.relu,
padding="SAME"):
with slim.arg_scope(
[slim.conv2d, slim.fully_connected],
weights_initializer=tf.truncated_normal_initializer(
dtype=tf.float32, stddev=0.01),
weights_regularizer=None,
biases_initializer=tf.constant_initializer(0.0),
):
with slim.arg_scope(
[slim.batch_norm],
decay=0.9,
scale=False,
epsilon=1e-3,
is_training=is_training,
zero_debias_moving_mean=True,
):
x = slim.conv2d(x, 24, [3, 3], normalizer_fn=tf.nn.lrn)
x = slim.max_pool2d(x, [2, 2])
x = slim.conv2d(x, 96, [3, 3], normalizer_fn=tf.nn.lrn)
x = slim.max_pool2d(x, [2, 2])
x = slim.conv2d(x, 192, [3, 3])
x = slim.conv2d(x, 192, [3, 3])
x = slim.conv2d(x, 96, [3, 3])
x = slim.max_pool2d(x, [2, 2])
# x = slim.dropout(x,is_training=is_training, keep_prob=0.9)
x = slim.flatten(x)
x = slim.fully_connected(x, 1024, activation_fn=None)
x = tflearn.relu(x)
x = slim.fully_connected(x, 1024, activation_fn=None)
x = tflearn.relu(x)
logits = slim.fully_connected(x,
class_num,
activation_fn=None)
return logits
def SepConv(x, C_out, kernel_size, stride):
x = tflearn.relu(x)
C_in = x.get_shape()[-1].value
x = slim.separable_convolution2d(x,
C_in,
kernel_size,
depth_multiplier=1,
stride=stride)
x = slim.batch_norm(x)
x = slim.separable_convolution2d(x, C_out, kernel_size, depth_multiplier=1)
x = slim.batch_norm(x)
return x
def build_network(self, num_classes, input_shape, model):
network = tflearn.input_data(
shape=[None, input_shape[0], input_shape[1], input_shape[2]])
if model == 'DeepFace':
conv_1 = tflearn.relu(
tflearn.conv_2d(network,
32,
11,
strides=1,
padding='VALID',
name='Conv2d_1'))
maxpool_1 = tflearn.max_pool_2d(conv_1,
3,
strides=2,
padding='VALID',
name='MaxPool_1')
conv_2 = tflearn.relu(
tflearn.conv_2d(maxpool_1,
32,
9,
strides=1,
padding='VALID',
name='Conv2d_2'))
local_1 = tflearn.relu(self.local(conv_2, 16, 9, 1, 'Local_1'))
local_2 = tflearn.relu(self.local(local_1, 16, 7, 1, 'Local_2'))
local_3 = tflearn.relu(self.local(local_2, 16, 5, 1, 'Local_3'))
flatterned = tflearn.flatten(local_3)
full_1 = tflearn.dropout(
tflearn.relu(
tflearn.fully_connected(flatterned,
4096,
name='Fully_Connected_1')), 0.5)
output = tflearn.fully_connected(full_1,
num_classes,
activation='softmax',
name='Output')
elif model == 'Song':
conv_1 = tflearn.relu(
tflearn.conv_2d(network,
64,
5,
strides=1,
padding='VALID',
name='Conv_1'))
maxpool_1 = tflearn.max_pool_2d(conv_1,
3,
strides=2,
padding='VALID',
name='MaxPool_1')
conv_2 = tflearn.relu(
tflearn.conv_2d(maxpool_1,
64,
5,
strides=1,
padding='VALID',
name='Conv_2'))
maxpool_2 = tflearn.max_pool_2d(conv_2,
3,
strides=2,
padding='VALID',
name='MaxPool_2')
local_1 = tflearn.dropout(
tflearn.relu(self.local(maxpool_2, 32, 3, 1, 'Local_1')), 1)
local_2 = tflearn.dropout(
tflearn.relu(self.local(local_1, 32, 3, 1, 'Local_2')), 1)
flatterned = tflearn.flatten(local_2)
output = tflearn.fully_connected(flatterned,
num_classes,
activation='softmax',
name='Output')
else:
conv_1 = tflearn.relu(
tflearn.conv_2d(network,
64,
7,
strides=2,
bias=True,
padding='VALID',
name='Conv2d_1'))
maxpool_1 = tflearn.batch_normalization(
tflearn.max_pool_2d(conv_1,
3,
strides=2,
padding='VALID',
name='MaxPool_1'))
conv_2a = tflearn.relu(
tflearn.conv_2d(maxpool_1,
96,
1,
strides=1,
padding='VALID',
name='Conv_2a_FX1'))
maxpool_2a = tflearn.max_pool_2d(maxpool_1,
3,
strides=1,
padding='VALID',
name='MaxPool_2a_FX1')
conv_2b = tflearn.relu(
tflearn.conv_2d(conv_2a,
208,
3,
strides=1,
padding='VALID',
name='Conv_2b_FX1'))
conv_2c = tflearn.relu(
tflearn.conv_2d(maxpool_2a,
64,
1,
strides=1,
padding='VALID',
name='Conv_2c_FX1'))
FX1_out = tflearn.merge([conv_2b, conv_2c],
mode='concat',
axis=3,
name='FX1_out')
conv_3a = tflearn.relu(
tflearn.conv_2d(FX1_out,
96,
1,
strides=1,
padding='VALID',
name='Conv_3a_FX2'))
maxpool_3a = tflearn.max_pool_2d(FX1_out,
3,
strides=1,
padding='VALID',
name='MaxPool_3a_FX2')
conv_3b = tflearn.relu(
tflearn.conv_2d(conv_3a,
208,
3,
strides=1,
padding='VALID',
name='Conv_3b_FX2'))
conv_3c = tflearn.relu(
tflearn.conv_2d(maxpool_3a,
64,
1,
strides=1,
padding='VALID',
name='Conv_3c_FX2'))
FX2_out = tflearn.merge([conv_3b, conv_3c],
mode='concat',
axis=3,
name='FX2_out')
net = tflearn.flatten(FX2_out)
output = tflearn.fully_connected(net,
num_classes,
activation='softmax',
name='Output')
return tflearn.regression(output,
optimizer='Adam',
loss='categorical_crossentropy',
learning_rate=0.000001)
def build_network(self):
padding = 'SAME'
print(' ')
print('----------------- Building CNN -----------------')
print(' ')
self.network = tflearn.input_data(
shape=[None, SIZE_FACE, SIZE_FACE, 1])
conv_1 = tflearn.relu(
conv_2d(self.network,
96,
3,
strides=1,
bias=True,
padding=padding,
activation=None,
name='Conv_1'))
maxpool_1 = tflearn.max_pool_2d(conv_1,
3,
strides=2,
padding=padding,
name='MaxPool_1')
maxpool_1 = tflearn.batch_normalization(maxpool_1)
conv_2 = tflearn.relu(
conv_2d(maxpool_1,
108,
2,
strides=1,
padding=padding,
name='Conv_2'))
maxpool_2 = tflearn.max_pool_2d(conv_2,
2,
strides=1,
padding=padding,
name='MaxPool_2')
maxpool_2 = tflearn.batch_normalization(maxpool_2)
conv_3 = tflearn.relu(
conv_2d(maxpool_2,
208,
2,
strides=1,
padding=padding,
name='Conv_3'))
conv_4 = tflearn.relu(
conv_2d(conv_3, 64, 2, strides=1, padding=padding, name='Conv_4'))
maxpool_3 = tflearn.max_pool_2d(conv_4,
2,
strides=1,
padding=padding,
name='MaxPool_3')
maxpool_3 = tflearn.batch_normalization(maxpool_3)
net = tflearn.flatten(maxpool_3, name='Net')
net = tflearn.dropout(net, 0.1)
final_1 = tflearn.fully_connected(net, 512, activation='relu')
final_1 = tflearn.dropout(final_1, 0.5)
final_2 = tflearn.fully_connected(final_1, 256, activation='relu')
final_2 = tflearn.dropout(final_2, 0.5)
Loss = tflearn.fully_connected(final_2,
7,
activation='softmax',
name='Total_loss')
self.network = tflearn.regression(Loss,
optimizer='Adam',
loss='categorical_crossentropy',
learning_rate=0.0001)
self.model = tflearn.DNN(self.network,
tensorboard_verbose=0,
tensorboard_dir=os.getcwd() + '/checkpoint',
checkpoint_path='./data/' +
'/emotion_recognition',
max_checkpoints=None)
#self.model = tflearn.DNN(self.network)
self.load_model()
prop_statement_input_4 = tflearn.input_data([None, prop_statement_time_steps, input_dim], name='prop_statement_input_4')
prop_hyps_input_4 = tflearn.input_data([None, prop_hyps_time_steps, input_dim], name='prop_hyps_input_4')
prop_statement_input_5 = tflearn.input_data([None, prop_statement_time_steps, input_dim], name='prop_statement_input_5')
prop_hyps_input_5 = tflearn.input_data([None, prop_hyps_time_steps, input_dim], name='prop_hyps_input_5')
with tf.variable_scope('prop') as scope:
prop_net_1 = prop_net(prop_statement_input_1, prop_hyps_input_1)
scope.reuse_variables()
prop_net_2 = prop_net(prop_statement_input_2, prop_hyps_input_2)
prop_net_3 = prop_net(prop_statement_input_3, prop_hyps_input_3)
prop_net_4 = prop_net(prop_statement_input_4, prop_hyps_input_4)
prop_net_5 = prop_net(prop_statement_input_5, prop_hyps_input_5)
main_net = tflearn.fully_connected(main_net, output_dim, weights_init='xavier')
main_net = tflearn.relu(main_net)
main_net = tflearn.fully_connected(main_net, output_dim, weights_init='xavier')
main_net = tflearn.relu(main_net)
main_net = tflearn.dropout(main_net, p)
prop_net_1 = tflearn.fully_connected(prop_net_1, output_dim, weights_init='xavier')
prop_net_1 = tflearn.relu(prop_net_1)
prop_net_1 = tflearn.dropout(prop_net_1, p)
prop_net_1 = tflearn.fully_connected(prop_net_1, output_dim, weights_init='xavier')
prop_net_1 = tflearn.relu(prop_net_1)
prop_net_1 = tflearn.dropout(prop_net_1, p)
prop_net_2 = tflearn.fully_connected(prop_net_2, output_dim, weights_init='xavier')
prop_net_2 = tflearn.relu(prop_net_2)
prop_net_2 = tflearn.dropout(prop_net_2, p)
prop_net_2 = tflearn.fully_connected(prop_net_2, output_dim, weights_init='xavier')
number_features = data.shape[1] number_examples = data.shape[0] data_prep = DataPreprocessing() data_prep.add_featurewise_zero_center() #data_prep.add_featurewise_stdnorm() # Build neural network net = tflearn.input_data(shape=[None, number_features], data_preprocessing=data_prep) # 1 fully connected net = tflearn.fully_connected(net, number_hidden) #tflearn.add_weights_regularizer(net, loss='L2') net = tflearn.batch_normalization(net) net = tflearn.relu(net) # 2 net = tflearn.fully_connected(net, number_hidden) #tflearn.add_weights_regularizer(net, loss='L2') net = tflearn.relu(net) # 3 net = tflearn.fully_connected(net, number_hidden) #tflearn.add_weights_regularizer(net, loss='L2') net = tflearn.batch_normalization(net) net = tflearn.relu(net) # 4
def ReLUConvBN(x, C_out):
x = tflearn.relu(x)
x = slim.conv2d(x, C_out, [1, 1])
x = slim.batch_norm(x)
return x
def my_model(X, y):
y = tf.one_hot(y, 10)
network = tflearn.conv_2d(X,
nb_filter=32,
filter_size=3,
strides=1,
activation="linear",
padding="valid",
name="conv1",
weight_decay=0.01)
network = tflearn.batch_normalization(network, name="bn1")
network = tflearn.relu(network)
network = tflearn.max_pool_2d(network,
kernel_size=2,
strides=2,
padding="same",
name="pool1")
network = tflearn.conv_2d(network,
nb_filter=64,
filter_size=3,
strides=1,
activation="linear",
padding="valid",
name="conv2",
weight_decay=0.01)
network = tflearn.batch_normalization(network, name="bn2")
network = tflearn.relu(network)
network = tflearn.max_pool_2d(network,
kernel_size=2,
strides=2,
padding="same",
name="pool2")
network = tflearn.flatten(network, name="flat1")
network = tflearn.fully_connected(network,
1024,
activation="linear",
name="fc1",
weight_decay=0.01)
network = tflearn.batch_normalization(network, name="bn2")
network = tflearn.relu(network)
network = tflearn.fully_connected(network,
1024,
activation="linear",
name="fc2",
weight_decay=0.01)
network = tflearn.batch_normalization(network, name="bn3")
network = tflearn.relu(network)
logits = tflearn.fully_connected(network,
10,
activation="softmax",
name="output",
weight_decay=0.01)
loss = tflearn.categorical_crossentropy(logits, y)
train_op = tflearn.Adam(0.0001, 0.9)().minimize(loss)
return logits, loss, train_op
