def get_resnet(x_tensor, reuse, is_training, x_batch_size):
with tf.variable_scope('resnet', reuse=reuse):
with slim.arg_scope(resnet_v2.resnet_arg_scope()):
resnet, end_points = resnet_v2.resnet_v2_50(
x_tensor,
global_pool=False,
is_training=is_training,
reuse=reuse,
output_stride=32)
global_pool = tf.reduce_mean(resnet, [1, 2])
with tf.variable_scope('fc'):
global_pool = slim.fully_connected(global_pool,
2048,
scope='fc/fc_1')
global_pool = slim.fully_connected(global_pool,
1024,
scope='fc/fc_2')
global_pool = slim.fully_connected(global_pool,
512,
scope='fc/fc_3')
theta = output_layer(global_pool, (grid_h + 1) * (grid_w + 1) * 2)
with tf.name_scope('gen_theta'):
id2_loss = tf.reduce_mean(tf.abs(theta)) * id_mul
return theta, id2_loss, id2_loss
def get_resnet(x_tensor, reuse, is_training, x_batch_size):
with tf.variable_scope('resnet', reuse=reuse):
with slim.arg_scope(resnet_v2.resnet_arg_scope()):
resnet, end_points = resnet_v2.resnet_v2_50(
x_tensor,
global_pool=False,
is_training=is_training,
reuse=reuse,
output_stride=32)
global_pool = tf.reduce_mean(resnet, [1, 2])
with tf.variable_scope('fc'):
theta = output_layer(global_pool, 8)
with tf.name_scope('gen_theta'):
eyes = tf.constant([1, 0, 0, 0, 1, 0, 0, 0, 1],
dtype=tf.float32) #, trainable=False)
eyes = tf.reshape(tf.tile(eyes, [x_batch_size]),
[x_batch_size, -1])
ones = tf.zeros(shape=[x_batch_size, 1], dtype=tf.float32)
id_loss = tf.reduce_mean(tf.abs(theta)) * id_mul
theta = tf.concat([theta, ones], 1)
theta = theta + eyes
return theta, id_loss
def reduce_layer(input):
with tf.variable_scope('reduce_layer'):
with tf.variable_scope('conv0'):
conv0_ = conv_bn_relu_layer(input, [1, 1, 2048, 512], 1)
with tf.variable_scope('conv1_0'):
conv1_0_ = conv_bn_relu_layer2(conv0_, [1, 16, 512, 512], [1, 1])
with tf.variable_scope('conv2_0'):
conv2_0_ = conv_bn_relu_layer2(conv1_0_, [9, 1, 512, 512], [1, 1])
with tf.variable_scope('conv1_1'):
conv1_1_ = conv_bn_relu_layer2(conv0_, [9, 1, 512, 512], [1, 1])
with tf.variable_scope('conv2_1'):
conv2_1_ = conv_bn_relu_layer2(conv1_1_, [1, 16, 512, 512], [1, 1])
with tf.variable_scope('conv2'):
conv2_ = conv2_0_ + conv2_1_
with tf.variable_scope('conv3'):
conv3_ = conv_bn_relu_layer(conv2_, [1, 1, 512, 128], 1)
with tf.variable_scope('conv4'):
conv4_ = conv_bn_relu_layer(conv3_, [1, 1, 128, 32], 1)
with tf.variable_scope('fc'):
out = output_layer(tf.reshape(conv4_, [batch_size, 32]), 8)
return out