def fc_layers(self, trainable):
# fc1
with tf.name_scope('fc1') as scope:
shape = int(np.prod(self.pool5.get_shape()[1:]))
fc1w = tf.Variable(tf.truncated_normal([shape, 4096],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
fc1b = tf.Variable(tf.constant(1.0, shape=[4096],
dtype=tf.float32),
trainable=trainable,
name='biases')
pool5_flat = tf.reshape(self.pool5, [-1, shape])
fc1l = tf.nn.bias_add(tf.matmul(pool5_flat, fc1w), fc1b)
self.fc1 = tf.nn.relu(fc1l)
self.after_relus.append(self.fc1)
self.parameters += [fc1w, fc1b]
# fc2
with tf.name_scope('fc2') as scope:
fc2w = tf.Variable(tf.truncated_normal([4096, 4096],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
fc2b = tf.Variable(tf.constant(1.0, shape=[4096],
dtype=tf.float32),
trainable=trainable,
name='biases')
fc2l = tf.nn.bias_add(tf.matmul(self.fc1, fc2w), fc2b)
self.fc2 = tf.nn.relu(fc2l)
self.after_relus.append(self.fc2)
self.parameters += [fc2w, fc2b]
if self.stop_at_fc2:
return
# fc3
with tf.name_scope('fc3') as scope:
fc3w = tf.Variable(tf.truncated_normal([4096, 1000],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
fc3b = tf.Variable(tf.constant(1.0, shape=[1000],
dtype=tf.float32),
trainable=trainable,
name='biases')
self.fc3l = tf.nn.bias_add(tf.matmul(self.fc2, fc3w), fc3b)
self.parameters += [fc3w, fc3b]
def dense_layer_ops(X, num_X, num_Y, config):
W = tf.Variable(tf.truncated_normal([num_X, num_Y],
mean=0.0,
stddev=np.float32(
config.var_init_stddev)),
name='W')
B = tf.Variable(tf.constant(value=np.float32(config.bias_init),
dtype=tf.float32,
shape=[num_Y]),
name='B')
Y = tf.add(tf.matmul(X, W), B, name='Y')
return W, B, Y
def get_reg_term(config, vars_to_reg):
reg_term = tf.constant(value=0.0, dtype=tf.float32)
l2reg = np.float32(config.l2reg)
l1reg = np.float32(config.l1reg)
if (l2reg > 0.0 or l1reg > 0.0) and len(vars_to_reg) > 0:
for x_var in vars_to_reg:
if l2reg > 0.0:
x_squared = x_var * x_var
l2norm = tf.reduce_sum(x_squared)
reg_term += l2reg * l2norm
if l1reg > 0.0:
x_abs = tf.abs(x_var)
l1norm = tf.reduce_sum(x_abs)
reg_term += l1reg * l1norm
return reg_term
def convlayers(self, trainable):
self.parameters = []
images = self.imgs
# conv1_1
with tf.name_scope('conv1_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 3, 64],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(images, kernel, [1, 1, 1, 1], padding='SAME')
self.conv_input_to_conv1_1 = conv
biases = tf.Variable(tf.constant(0.0, shape=[64],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv1_1 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv1_1)
self.parameters += [kernel, biases]
# conv1_2
with tf.name_scope('conv1_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 64, 64],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.conv1_1,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv1_2 = conv
biases = tf.Variable(tf.constant(0.0, shape=[64],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv1_2 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv1_2)
self.parameters += [kernel, biases]
# pool1
self.pool1 = tf.nn.max_pool(self.conv1_2,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool1')
# conv2_1
with tf.name_scope('conv2_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 64, 128],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.pool1,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv2_1 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[128],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv2_1 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv2_1)
self.parameters += [kernel, biases]
# conv2_2
with tf.name_scope('conv2_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 128, 128],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.conv2_1,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv2_2 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[128],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv2_2 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv2_2)
self.parameters += [kernel, biases]
# pool2
self.pool2 = tf.nn.max_pool(self.conv2_2,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool2')
# conv3_1
with tf.name_scope('conv3_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 128, 256],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.pool2,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv3_1 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[256],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv3_1 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv3_1)
self.parameters += [kernel, biases]
# conv3_2
with tf.name_scope('conv3_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 256, 256],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.conv3_1,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv3_2 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[256],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv3_2 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv3_2)
self.parameters += [kernel, biases]
# conv3_3
with tf.name_scope('conv3_3') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 256, 256],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.conv3_2,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv3_3 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[256],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv3_3 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv3_3)
self.parameters += [kernel, biases]
# pool3
self.pool3 = tf.nn.max_pool(self.conv3_3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool3')
# conv4_1
with tf.name_scope('conv4_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 256, 512],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.pool3,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv4_1 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[512],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv4_1 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv4_1)
self.parameters += [kernel, biases]
# conv4_2
with tf.name_scope('conv4_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.conv4_1,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv4_2 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[512],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv4_2 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv4_2)
self.parameters += [kernel, biases]
# conv4_3
with tf.name_scope('conv4_3') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.conv4_2,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv4_3 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[512],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv4_3 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv4_3)
self.parameters += [kernel, biases]
# pool4
self.pool4 = tf.nn.max_pool(self.conv4_3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool4')
# conv5_1
with tf.name_scope('conv5_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.pool4,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv5_1 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[512],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv5_1 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv5_1)
self.parameters += [kernel, biases]
# conv5_2
with tf.name_scope('conv5_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.conv5_1,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv5_2 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[512],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv5_2 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv5_2)
self.parameters += [kernel, biases]
# conv5_3
with tf.name_scope('conv5_3') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512],
dtype=tf.float32,
stddev=1e-1),
name='weights',
trainable=trainable)
conv = tf.nn.conv2d(self.conv5_2,
kernel, [1, 1, 1, 1],
padding='SAME')
self.conv_input_to_conv5_3 = conv
biases = tf.Variable(tf.constant(0.0,
shape=[512],
dtype=tf.float32),
trainable=trainable,
name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv5_3 = tf.nn.relu(out, name=scope)
self.after_relus.append(self.conv5_3)
self.parameters += [kernel, biases]
# pool5
self.pool5 = tf.nn.max_pool(self.conv5_3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool4')