def create_conv_net(x,
keep_prob,
channels,
n_class,
layers=3,
features_root=16,
filter_size=3,
pool_size=2,
summaries=True):
nx = tf.shape(x)[1]
ny = tf.shape(x)[2]
x_image = tf.reshape(x, tf.stack([-1, nx, ny, channels]))
in_node = x_image
batch_size = tf.shape(x_image)[0]
weights = []
biases = []
convs = []
pools = OrderedDict()
deconv = OrderedDict()
dw_h_convs = OrderedDict()
up_h_convs = OrderedDict()
in_size = 1000
size = in_size
# down layers
for layer in range(0, layers):
features = 2**layer * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
if layer == 0:
w1 = weight_variable(
[filter_size, filter_size, channels, features], stddev)
else:
w1 = weight_variable(
[filter_size, filter_size, features // 2, features], stddev)
w2 = weight_variable([filter_size, filter_size, features, features],
stddev)
b1 = bias_variable([features])
b2 = bias_variable([features])
conv1 = conv2d(in_node, w1, keep_prob)
tmp_h_conv = tf.nn.relu(conv1 + b1)
conv2 = conv2d(tmp_h_conv, w2, keep_prob)
dw_h_convs[layer] = tf.nn.relu(conv2 + b2)
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size -= 4
if layer < layers - 1:
pools[layer] = max_pool(dw_h_convs[layer], pool_size)
in_node = pools[layer]
size /= 2
in_node = dw_h_convs[layers - 1]
# up layers
for layer in range(layers - 2, -1, -1):
features = 2**(layer + 1) * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
wd = weight_variable_devonc(
[pool_size, pool_size, features // 2, features], stddev)
bd = bias_variable([features // 2])
h_deconv = tf.nn.relu(deconv2d(in_node, wd, pool_size) + bd)
h_deconv_concat = crop_and_concat(dw_h_convs[layer], h_deconv)
deconv[layer] = h_deconv_concat
w1 = weight_variable(
[filter_size, filter_size, features, features // 2], stddev)
w2 = weight_variable(
[filter_size, filter_size, features // 2, features // 2], stddev)
b1 = bias_variable([features // 2])
b2 = bias_variable([features // 2])
conv1 = conv2d(h_deconv_concat, w1, keep_prob)
h_conv = tf.nn.relu(conv1 + b1)
conv2 = conv2d(h_conv, w2, keep_prob)
in_node = tf.nn.relu(conv2 + b2)
up_h_convs[layer] = in_node
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size *= 2
size -= 4
# Output Map
weight = weight_variable([1, 1, features_root, n_class], stddev)
bias = bias_variable([n_class])
conv = conv2d(in_node, weight, tf.constant(1.0))
output_map = tf.nn.relu(conv + bias)
up_h_convs["out"] = output_map
if summaries:
for i, (c1, c2) in enumerate(convs):
tf.summary.image('summary_conv_%02d_01' % i, get_image_summary(c1))
tf.summary.image('summary_conv_%02d_02' % i, get_image_summary(c2))
for k in pools.keys():
tf.summary.image('summary_pool_%02d' % k,
get_image_summary(pools[k]))
for k in deconv.keys():
tf.summary.image('summary_deconv_concat_%02d' % k,
get_image_summary(deconv[k]))
for k in dw_h_convs.keys():
tf.summary.histogram("dw_convolution_%02d" % k + '/activations',
dw_h_convs[k])
for k in up_h_convs.keys():
tf.summary.histogram("up_convolution_%s" % k + '/activations',
up_h_convs[k])
variables = []
for w1, w2 in weights:
variables.append(w1)
variables.append(w2)
for b1, b2 in biases:
variables.append(b1)
variables.append(b2)
return output_map, variables, int(in_size - size)
def create_conv_net(x,
keep_prob,
channels,
n_class,
layers=3,
features_root=16,
filter_size=3,
pool_size=2,
summaries=True):
"""
Creates a new convolutional unet for the given parametrization.
:param x: input tensor, shape [?,nx,ny,channels]
:param keep_prob: dropout probability tensor
:param channels: number of channels in the input image
:param n_class: number of output labels
:param layers: number of layers in the net
:param features_root: number of features in the first layer
:param filter_size: size of the convolution filter
:param pool_size: size of the max pooling operation
:param summaries: Flag if summaries should be created
"""
logging.info(
"Layers {layers}, features {features}, filter size {filter_size}x{filter_size}, pool size: {pool_size}x{pool_size}"
.format(layers=layers,
features=features_root,
filter_size=filter_size,
pool_size=pool_size))
# Placeholder for the input image
nx = tf.shape(x)[1]
ny = tf.shape(x)[2]
x_image = tf.reshape(x, tf.stack([-1, nx, ny, channels]))
in_node = x_image
batch_size = tf.shape(x_image)[0]
weights = []
biases = []
convs = []
pools = OrderedDict()
deconv = OrderedDict()
dw_h_convs = OrderedDict()
up_h_convs = OrderedDict()
in_size = 1000
size = in_size
# down layers
for layer in range(0, layers):
features = 2**layer * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
if layer == 0:
w1 = weight_variable(
[filter_size, filter_size, channels, features], stddev)
else:
w1 = weight_variable(
[filter_size, filter_size, features // 2, features], stddev)
w2 = weight_variable([filter_size, filter_size, features, features],
stddev)
b1 = bias_variable([features])
b2 = bias_variable([features])
conv1 = conv2d(in_node, w1, keep_prob)
tmp_h_conv = tf.nn.relu(conv1 + b1)
conv2 = conv2d(tmp_h_conv, w2, keep_prob)
dw_h_convs[layer] = tf.nn.relu(conv2 + b2)
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size -= 4
if layer < layers - 1:
pools[layer] = max_pool(dw_h_convs[layer], pool_size)
in_node = pools[layer]
size /= 2
in_node = dw_h_convs[layers - 1]
# up layers
for layer in range(layers - 2, -1, -1):
features = 2**(layer + 1) * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
wd = weight_variable_devonc(
[pool_size, pool_size, features // 2, features], stddev)
bd = bias_variable([features // 2])
h_deconv = tf.nn.relu(deconv2d(in_node, wd, pool_size) + bd)
h_deconv_concat = crop_and_concat(dw_h_convs[layer], h_deconv)
deconv[layer] = h_deconv_concat
w1 = weight_variable(
[filter_size, filter_size, features, features // 2], stddev)
w2 = weight_variable(
[filter_size, filter_size, features // 2, features // 2], stddev)
b1 = bias_variable([features // 2])
b2 = bias_variable([features // 2])
conv1 = conv2d(h_deconv_concat, w1, keep_prob)
h_conv = tf.nn.relu(conv1 + b1)
conv2 = conv2d(h_conv, w2, keep_prob)
in_node = tf.nn.relu(conv2 + b2)
up_h_convs[layer] = in_node
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size *= 2
size -= 4
# Output Map
weight = weight_variable([1, 1, features_root, n_class], stddev)
bias = bias_variable([n_class])
conv = conv2d(in_node, weight, tf.constant(1.0))
output_map = tf.nn.relu(conv + bias)
up_h_convs["out"] = output_map
if summaries:
for i, (c1, c2) in enumerate(convs):
tf.summary.image('summary_conv_%02d_01' % i, get_image_summary(c1))
tf.summary.image('summary_conv_%02d_02' % i, get_image_summary(c2))
for k in pools.keys():
tf.summary.image('summary_pool_%02d' % k,
get_image_summary(pools[k]))
for k in deconv.keys():
tf.summary.image('summary_deconv_concat_%02d' % k,
get_image_summary(deconv[k]))
for k in dw_h_convs.keys():
tf.summary.histogram("dw_convolution_%02d" % k + '/activations',
dw_h_convs[k])
for k in up_h_convs.keys():
tf.summary.histogram("up_convolution_%s" % k + '/activations',
up_h_convs[k])
variables = []
for w1, w2 in weights:
variables.append(w1)
variables.append(w2)
for b1, b2 in biases:
variables.append(b1)
variables.append(b2)
return output_map, variables, int(in_size - size)
def unet(
x,
is_training,
keep_prob=1,
channels=1,
n_class=1,
layers=3,
features_root=64,
filter_size=3,
pool_size=2,
summaries=False,
):
"""
Creates a new convolutional unet for the given parametrization.
:param x: input tensor, shape [?,nx,ny,channels]
:param keep_prob: dropout probability tensor
:param channels: number of channels in the input image
:param n_class: number of output labels
:param layers: number of layers in the net
:param features_root: number of features in the first layer
:param filter_size: size of the convolution filter
:param pool_size: size of the max pooling operation
:param summaries: Flag if summaries should be created
"""
with tf.device('/gpu:0'):
logging.info(
'Layers {layers}, features {features}, filter size {filter_size}x{filter_size}, pool size: {pool_size}x{pool_size}'
.format(layers=layers,
features=features_root,
filter_size=filter_size,
pool_size=pool_size))
# Placeholder for the input image
nx = tf.shape(x)[1]
ny = tf.shape(x)[2]
x_image = tf.reshape(x, tf.stack([-1, nx, ny, channels]))
in_node = x_image
batch_size = tf.shape(x_image)[0]
weights = []
biases = []
convs = []
pools = OrderedDict()
deconv = OrderedDict()
dw_h_convs = OrderedDict()
up_h_convs = OrderedDict()
in_size = 1000
size = in_size
# down layers
for layer in range(0, layers):
features = 2**layer * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
if layer == 0:
w1 = tf.get_variable(
'down_conv_00_w1',
[filter_size, filter_size, channels, features],
initializer=tf.random_normal_initializer(stddev=stddev))
else:
w1 = tf.get_variable(
'down_conv_%02d_w1' % (layer + 1),
[filter_size, filter_size, features // 2, features],
initializer=tf.random_normal_initializer(stddev=stddev))
w2 = tf.get_variable(
'down_conv_%02d_w2' % (layer + 1),
[filter_size, filter_size, features, features],
initializer=tf.random_normal_initializer(stddev=stddev))
b1 = tf.get_variable('conv_%02d_b1' % (layer + 1), [features],
initializer=tf.constant_initializer(0.1))
b2 = tf.get_variable('conv_%02d_b2' % (layer + 1), [features],
initializer=tf.constant_initializer(0.1))
conv1 = conv2d(in_node, w1, keep_prob)
print(conv1.get_shape())
conv1 = batch_norm_wrapper(conv1, is_training)
tmp_h_conv = tf.nn.relu(conv1 + b1)
conv2 = conv2d(tmp_h_conv, w2, keep_prob)
conv2 = batch_norm_wrapper(conv2, is_training)
dw_h_convs[layer] = tf.nn.relu(conv2 + b2)
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
if layer < layers - 1:
pools[layer] = max_pool(dw_h_convs[layer], pool_size)
in_node = pools[layer]
in_node = dw_h_convs[layers - 1]
# up layers
for layer in range(layers - 2, -1, -1):
features = 2**(layer + 1) * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
# wd = weight_variable_devonc([pool_size, pool_size, features//2, features], stddev)
wd = tf.get_variable(
'up_conv_%02d_wd' % (layer + 1),
[pool_size, pool_size, features // 2, features],
initializer=tf.random_normal_initializer(stddev=stddev))
# bd = bias_variable([features//2])
bd = tf.get_variable('up_conv_%02d_bd' % (layer + 1),
[features // 2],
initializer=tf.constant_initializer(0.1))
h_deconv = tf.nn.relu(deconv2d(in_node, wd, pool_size) + bd)
h_deconv_concat = crop_and_concat(dw_h_convs[layer], h_deconv)
deconv[layer] = h_deconv_concat
# w1 = weight_variable([filter_size, filter_size, features, features//2], stddev)
w1 = tf.get_variable(
'up_conv_%02d_w1' % (layer + 1),
[filter_size, filter_size, features, features // 2],
initializer=tf.random_normal_initializer(stddev=stddev))
# w2 = weight_variable([filter_size, filter_size, features//2, features//2], stddev)
w2 = tf.get_variable(
'up_conv_%02d_w2' % (layer + 1),
[filter_size, filter_size, features // 2, features // 2],
initializer=tf.random_normal_initializer(stddev=stddev))
# b1 = bias_variable([features//2])
b1 = tf.get_variable('up_conv_%02d_b1' % (layer + 1),
[features // 2],
initializer=tf.constant_initializer(0.1))
# b2 = bias_variable([features//2])
b2 = tf.get_variable('up_conv_%02d_b2' % (layer + 1),
[features // 2],
initializer=tf.constant_initializer(0.1))
conv1 = conv2d(h_deconv_concat, w1, keep_prob)
conv1 = batch_norm_wrapper(conv1, is_training)
h_conv = tf.nn.relu(conv1 + b1)
conv2 = conv2d(h_conv, w2, keep_prob)
conv2 = batch_norm_wrapper(conv2, is_training)
in_node = tf.nn.relu(conv2 + b2)
up_h_convs[layer] = in_node
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
# size *= 2
# size -= 4
# Output Map
# weight = weight_variable([1, 1, features_root, n_class], stddev)
weight = tf.get_variable(
'weight', [1, 1, features_root, n_class],
initializer=tf.random_normal_initializer(stddev=stddev))
# bias = bias_variable([n_class])
bias = tf.get_variable('bias', [n_class],
initializer=tf.constant_initializer(0.1))
conv = conv2d(in_node, weight, tf.constant(1.0))
# conv = batch_norm_wrapper(conv, is_training)
output_map = tf.nn.relu(conv + bias)
# output_map = tf.add(output_map, x_image)
up_h_convs['out'] = output_map
if summaries:
for (i, (c1, c2)) in enumerate(convs):
tf.summary.image('summary_conv_%02d_01' % i,
get_image_summary(c1))
tf.summary.image('summary_conv_%02d_02' % i,
get_image_summary(c2))
for k in pools.keys():
tf.summary.image('summary_pool_%02d' % k,
get_image_summary(pools[k]))
for k in deconv.keys():
tf.summary.image('summary_deconv_concat_%02d' % k,
get_image_summary(deconv[k]))
for k in dw_h_convs.keys():
tf.summary.histogram(
'dw_convolution_%02d' % k + '/activations', dw_h_convs[k])
for k in up_h_convs.keys():
tf.summary.histogram('up_convolution_%s' % k + '/activations',
up_h_convs[k])
variables = []
for (w1, w2) in weights:
variables.append(w1)
variables.append(w2)
for (b1, b2) in biases:
variables.append(b1)
variables.append(b2)
# return output_map, variables, int(in_size - size)
return (output_map, variables)
def create_conv_net(x, keep_prob, channels, n_class, unet_kwargs):
"""
Creates a new convolutional unet for the given parametrization.
:param x: input tensor, shape [?,nx,ny,channels]
:param keep_prob: dropout probability tensor
:param channels: number of channels in the input image
:param n_class: number of output labels
:param layers: number of layers in the net
:param features_root: number of features in the first layer
:param filter_size: size of the convolution filter
:param pool_size: size of the max pooling operation
"""
layers = unet_kwargs.pop('layers', 3)
features_root = unet_kwargs.pop('features_root', 16)
filter_size = unet_kwargs.pop('filter_size', 3)
pool_size = unet_kwargs.pop('pool_size', 2)
logging.info(
"Layers {layers}, features {features}, filter size {filter_size}x{filter_size}, pool size: {pool_size}x{pool_size}"
.format(layers=layers,
features=features_root,
filter_size=filter_size,
pool_size=pool_size))
# Placeholder for the input image
nx = tf.shape(x)[1]
ny = tf.shape(x)[2]
x_image = tf.reshape(x, tf.stack([-1, nx, ny, channels]))
in_node = x_image
batch_size = tf.shape(x_image)[0]
with tf.variable_scope('generator'):
pools = OrderedDict()
deconv = OrderedDict()
dw_h_convs = OrderedDict()
up_h_convs = OrderedDict()
in_size = 1000
size = in_size
# down layers
for layer in range(0, layers):
features = 2**layer * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
if layer == 0:
w1 = weight_variable(
[filter_size, filter_size, channels, features], stddev)
else:
w1 = weight_variable(
[filter_size, filter_size, features // 2, features],
stddev)
w2 = weight_variable(
[filter_size, filter_size, features, features], stddev)
b1 = bias_variable([features])
b2 = bias_variable([features])
conv1 = conv2d(in_node, w1, keep_prob)
tmp_h_conv = tf.nn.relu(conv1 + b1)
conv2 = conv2d(tmp_h_conv, w2, keep_prob)
dw_h_convs[layer] = tf.nn.relu(conv2 + b2)
size -= 4
if layer < layers - 1:
pools[layer] = max_pool(dw_h_convs[layer], pool_size)
in_node = pools[layer]
size /= 2
in_node = dw_h_convs[layers - 1]
# up layers
for layer in range(layers - 2, -1, -1):
features = 2**(layer + 1) * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
wd = weight_variable_devonc(
[pool_size, pool_size, features // 2, features], stddev)
bd = bias_variable([features // 2])
h_deconv = tf.nn.relu(deconv2d(in_node, wd, pool_size) + bd)
h_deconv_concat = crop_and_concat(dw_h_convs[layer], h_deconv)
deconv[layer] = h_deconv_concat
w1 = weight_variable(
[filter_size, filter_size, features, features // 2], stddev)
w2 = weight_variable(
[filter_size, filter_size, features // 2, features // 2],
stddev)
b1 = bias_variable([features // 2])
b2 = bias_variable([features // 2])
conv1 = conv2d(h_deconv_concat, w1, keep_prob)
h_conv = tf.nn.relu(conv1 + b1)
conv2 = conv2d(h_conv, w2, keep_prob)
in_node = tf.nn.relu(conv2 + b2)
up_h_convs[layer] = in_node
size *= 2
size -= 4
# Output Map
weight = weight_variable([1, 1, features_root, n_class], stddev)
bias = bias_variable([n_class])
conv = conv2d(in_node, weight, tf.constant(1.0))
output_map = tf.nn.relu(conv + bias)
up_h_convs["out"] = output_map
return output_map, tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,
'generator'), int(in_size - size)
def create_conv_net(x, keep_prob, channels, n_class, layers=3, features_root=16, filter_size=3, pool_size=2,
summaries=True):
"""
Creates a new convolutional unet for the given parametrization.
:param x: input tensor, shape [?,nx,ny,channels]
:param keep_prob: dropout probability tensor
:param channels: number of channels in the input image
:param n_class: number of output labels
:param layers: number of layers in the net
:param features_root: number of features in the first layer
:param filter_size: size of the convolution filter
:param pool_size: size of the max pooling operation
:param summaries: Flag if summaries should be created
"""
logging.info(
"Layers {layers}, features {features}, filter size {filter_size}x{filter_size}, pool size: {pool_size}x{pool_size}".format(
layers=layers,
features=features_root,
filter_size=filter_size,
pool_size=pool_size))
# Placeholder for the input image
with tf.name_scope("preprocessing"):
nx = tf.shape(x)[1]
ny = tf.shape(x)[2]
x_image = tf.reshape(x, tf.stack([-1, nx, ny, channels]))
in_node = x_image
batch_size = tf.shape(x_image)[0]
weights = []
biases = []
convs = []
pools = OrderedDict()
deconv = OrderedDict()
dw_h_convs = OrderedDict()
up_h_convs = OrderedDict()
in_size = 1000
size = in_size
# down layers
for layer in range(0, layers):
with tf.name_scope("down_conv_{}".format(str(layer))):
features = 2 ** layer * features_root
stddev = np.sqrt(2 / (filter_size ** 2 * features))
if layer == 0:
w1 = weight_variable([filter_size, filter_size, channels, features], stddev, name="w1")
else:
w1 = weight_variable([filter_size, filter_size, features // 2, features], stddev, name="w1")
w2 = weight_variable([filter_size, filter_size, features, features], stddev, name="w2")
b1 = bias_variable([features], name="b1")
b2 = bias_variable([features], name="b2")
conv1 = conv2d(in_node, w1, b1, keep_prob)
tmp_h_conv = tf.nn.relu(conv1)
conv2 = conv2d(tmp_h_conv, w2, b2, keep_prob)
dw_h_convs[layer] = tf.nn.relu(conv2)
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size -= 4
if layer < layers - 1:
pools[layer] = max_pool(dw_h_convs[layer], pool_size)
in_node = pools[layer]
size /= 2
in_node = dw_h_convs[layers - 1]
# up layers
for layer in range(layers - 2, -1, -1):
with tf.name_scope("up_conv_{}".format(str(layer))):
features = 2 ** (layer + 1) * features_root
stddev = np.sqrt(2 / (filter_size ** 2 * features))
wd = weight_variable_devonc([pool_size, pool_size, features // 2, features], stddev, name="wd")
bd = bias_variable([features // 2], name="bd")
h_deconv = tf.nn.relu(deconv2d(in_node, wd, pool_size) + bd)
h_deconv_concat = crop_and_concat(dw_h_convs[layer], h_deconv)
deconv[layer] = h_deconv_concat
w1 = weight_variable([filter_size, filter_size, features, features // 2], stddev, name="w1")
w2 = weight_variable([filter_size, filter_size, features // 2, features // 2], stddev, name="w2")
b1 = bias_variable([features // 2], name="b1")
b2 = bias_variable([features // 2], name="b2")
conv1 = conv2d(h_deconv_concat, w1, b1, keep_prob)
h_conv = tf.nn.relu(conv1)
conv2 = conv2d(h_conv, w2, b2, keep_prob)
in_node = tf.nn.relu(conv2)
up_h_convs[layer] = in_node
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size *= 2
size -= 4
# Output Map
with tf.name_scope("output_map"):
weight = weight_variable([1, 1, features_root, n_class], stddev)
bias = bias_variable([n_class], name="bias")
conv = conv2d(in_node, weight, bias, tf.constant(1.0))
output_map = tf.nn.relu(conv)
up_h_convs["out"] = output_map
if summaries:
with tf.name_scope("summaries"):
for i, (c1, c2) in enumerate(convs):
tf.summary.image('summary_conv_%02d_01' % i, get_image_summary(c1))
tf.summary.image('summary_conv_%02d_02' % i, get_image_summary(c2))
for k in pools.keys():
tf.summary.image('summary_pool_%02d' % k, get_image_summary(pools[k]))
for k in deconv.keys():
tf.summary.image('summary_deconv_concat_%02d' % k, get_image_summary(deconv[k]))
for k in dw_h_convs.keys():
tf.summary.histogram("dw_convolution_%02d" % k + '/activations', dw_h_convs[k])
for k in up_h_convs.keys():
tf.summary.histogram("up_convolution_%s" % k + '/activations', up_h_convs[k])
variables = []
for w1, w2 in weights:
variables.append(w1)
variables.append(w2)
for b1, b2 in biases:
variables.append(b1)
variables.append(b2)
return output_map, variables, int(in_size - size)
def create_conv_net(x, keep_prob, channels, n_class, layers=3, features_root=16, filter_size=3, pool_size=2,
summaries=True):
"""
Creates a new convolutional unet for the given parametrization.
:param x: input tensor, shape [?,nx,ny,channels]
:param keep_prob: dropout probability tensor
:param channels: number of channels in the input image
:param n_class: number of output labels
:param layers: number of layers in the net
:param features_root: number of features in the first layer
:param filter_size: size of the convolution filter
:param pool_size: size of the max pooling operation
:param summaries: Flag if summaries should be created
"""
logging.info(
"Layers {layers}, features {features}, filter size {filter_size}x{filter_size}, pool size: {pool_size}x{pool_size}".format(
layers=layers,
features=features_root,
filter_size=filter_size,
pool_size=pool_size))
# Placeholder for the input image
with tf.name_scope("preprocessing"):
nx = tf.shape(x)[1]
ny = tf.shape(x)[2]
x_image = tf.reshape(x, tf.stack([-1, nx, ny, channels]))
in_node = x_image
batch_size = tf.shape(x_image)[0]
weights = []
biases = []
convs = []
pools = OrderedDict()
deconv = OrderedDict()
dw_h_convs = OrderedDict()
up_h_convs = OrderedDict()
in_size = 1000
size = in_size
# down layers
for layer in range(0, layers):
with tf.name_scope("down_conv_{}".format(str(layer))):
features = 2 ** layer * features_root #每层的特征为16,32,64, 是2的幂次数
stddev = np.sqrt(2 / (filter_size ** 2 * features))
if layer == 0:
w1 = weight_variable([filter_size, filter_size, channels, features], stddev, name="w1") #第一层直接channels变features [3,3,1,16]
else:
w1 = weight_variable([filter_size, filter_size, features // 2, features], stddev, name="w1") #第二层和第三层channels变两倍
w2 = weight_variable([filter_size, filter_size, features, features], stddev, name="w2") #w : [width, height, channels, kernel_nums]
b1 = bias_variable([features], name="b1") # x : [batch_size,image_height,image_width,channels]
b2 = bias_variable([features], name="b2") # b : [channels]
conv1 = conv2d(in_node, w1, b1, keep_prob) # [?,?,?,16]
tmp_h_conv = tf.nn.relu(conv1)
conv2 = conv2d(tmp_h_conv, w2, b2, keep_prob)
dw_h_convs[layer] = tf.nn.relu(conv2)
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size -= 2 * 2 * (filter_size // 2) # valid conv
if layer < layers - 1: #直到倒数第二个
pools[layer] = max_pool(dw_h_convs[layer], pool_size)
in_node = pools[layer]
size /= pool_size #总的图片大小变为一半
in_node = dw_h_convs[layers - 1] #使最后第一个层变为输入
# up layers
for layer in range(layers - 2, -1, -1):
with tf.name_scope("up_conv_{}".format(str(layer))):
features = 2 ** (layer + 1) * features_root #feature倒过来，64,32,16
stddev = np.sqrt(2 / (filter_size ** 2 * features))
wd = weight_variable_devonc([pool_size, pool_size, features // 2, features], stddev, name="wd") # [2,2,32,64]
bd = bias_variable([features // 2], name="bd") # 32
h_deconv = tf.nn.relu(deconv2d(in_node, wd, pool_size) + bd) # 下采样最后一个层来进行上采样，然后加上 下采样的倒数第二层
h_deconv_concat = crop_and_concat(dw_h_convs[layer], h_deconv) #此时layer为1,即倒数第二层
deconv[layer] = h_deconv_concat
w1 = weight_variable([filter_size, filter_size, features, features // 2], stddev, name="w1") # [3,3,64,32],这里才开始上采样，可以看到features // 2（变小）
w2 = weight_variable([filter_size, filter_size, features // 2, features // 2], stddev, name="w2") # [3,3,32,32]
b1 = bias_variable([features // 2], name="b1")
b2 = bias_variable([features // 2], name="b2")
conv1 = conv2d(h_deconv_concat, w1, b1, keep_prob)
h_conv = tf.nn.relu(conv1)
conv2 = conv2d(h_conv, w2, b2, keep_prob)
in_node = tf.nn.relu(conv2)
up_h_convs[layer] = in_node
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size *= pool_size
size -= 2 * 2 * (filter_size // 2) # valid conv
# Output Map
with tf.name_scope("output_map"):
weight = weight_variable([1, 1, features_root, n_class], stddev) #（1,1,16,2）
bias = bias_variable([n_class], name="bias")
conv = conv2d(in_node, weight, bias, tf.constant(1.0))
output_map = tf.nn.relu(conv)
up_h_convs["out"] = output_map
if summaries:
with tf.name_scope("summaries"):
for i, (c1, c2) in enumerate(convs):
tf.summary.image('summary_conv_%02d_01' % i, get_image_summary(c1))
tf.summary.image('summary_conv_%02d_02' % i, get_image_summary(c2))
for k in pools.keys():
tf.summary.image('summary_pool_%02d' % k, get_image_summary(pools[k]))
for k in deconv.keys():
tf.summary.image('summary_deconv_concat_%02d' % k, get_image_summary(deconv[k]))
for k in dw_h_convs.keys():
tf.summary.histogram("dw_convolution_%02d" % k + '/activations', dw_h_convs[k])
for k in up_h_convs.keys():
tf.summary.histogram("up_convolution_%s" % k + '/activations', up_h_convs[k])
variables = []
for w1, w2 in weights:
variables.append(w1)
variables.append(w2)
for b1, b2 in biases:
variables.append(b1)
variables.append(b2)
return output_map, variables, int(in_size - size)
def create_conv_net(x,
keep_prob,
channels,
n_class,
layers=3,
features_root=16,
filter_size=3,
pool_size=2,
summaries=True):
"""
Creates a new convolutional unet for the given parametrization.
:param x: input tensor, shape [?,nx,ny,channels]
:param keep_prob: dropout probability tensor
:param channels: number of channels in the input image
:param n_class: number of output labels
:param layers: number of layers in the net
:param features_root: number of features in the first layer
:param filter_size: size of the convolution filter
:param pool_size: size of the max pooling operation
:param summaries: Flag if summaries should be created
"""
logging.info(
"Layers {layers}, features {features}, filter size {filter_size}x{filter_size}, pool size: {pool_size}x{pool_size}"
.format(layers=layers,
features=features_root,
filter_size=filter_size,
pool_size=pool_size))
# Placeholder for the input image
nx = tf.shape(x)[1]
ny = tf.shape(x)[2]
x_image = tf.reshape(x, tf.stack([-1, nx, ny, channels]))
in_node = x_image
batch_size = tf.shape(x_image)[0]
weights = []
biases = []
convs = []
pools = OrderedDict()
deconv = OrderedDict()
dw_h_convs = OrderedDict()
up_h_convs = OrderedDict()
in_size = 1000
size = in_size
# find the number of pixels to suppress in one edge according to size of convolution filter
pixelEdge = (filter_size - 1) / 2
# compute the total number of pixel to remove at each time we use convolution filter
step = pixelEdge * 2
# Number of convolution in one layers (it is not a parameters)
nLayerofConvolution = 2
# Variable use to count the number of convolution and max-pooling into architecture
nConvFilter = 0
nMaxPooling = 0
with tf.name_scope('DOWN_LAYER') as scope:
# down layers
for layer in range(0, layers):
features = 2**layer * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
if layer == 0:
w1 = weight_variable(
[filter_size, filter_size, channels, features], stddev)
else:
w1 = weight_variable(
[filter_size, filter_size, features // 2, features],
stddev)
w2 = weight_variable(
[filter_size, filter_size, features, features], stddev)
b1 = bias_variable([features])
b2 = bias_variable([features])
conv1 = conv2d(in_node, w1, keep_prob)
tmp_h_conv = tf.nn.relu(conv1 + b1)
conv2 = conv2d(tmp_h_conv, w2, keep_prob)
dw_h_convs[layer] = tf.nn.relu(conv2 + b2)
nConvFilter += 2
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size = math.floor(size - (step * nLayerofConvolution))
if layer < layers - 1:
pools[layer] = max_pool(dw_h_convs[layer], pool_size)
in_node = pools[layer]
size = math.floor(size / pool_size)
nMaxPooling += 1
in_node = dw_h_convs[layers - 1]
with tf.name_scope('UP_LAYERS') as scope:
# up layers
for layer in range(layers - 2, -1, -1):
features = 2**(layer + 1) * features_root
stddev = np.sqrt(2 / (filter_size**2 * features))
wd = weight_variable_devonc(
[pool_size, pool_size, features // 2, features], stddev)
bd = bias_variable([features // 2])
h_deconv = tf.nn.relu(deconv2d(in_node, wd, pool_size) +
bd) #up-conv 2x2
h_deconv_concat = crop_and_concat(dw_h_convs[layer],
h_deconv) #copy and crop
deconv[layer] = h_deconv_concat
size = math.floor(size * pool_size)
nMaxPooling += 1
nConvFilter += 1
w1 = weight_variable(
[filter_size, filter_size, features, features // 2], stddev)
w2 = weight_variable(
[filter_size, filter_size, features // 2, features // 2],
stddev)
b1 = bias_variable([features // 2])
b2 = bias_variable([features // 2])
conv1 = conv2d(h_deconv_concat, w1, keep_prob) #conv
h_conv = tf.nn.relu(conv1 + b1) #relu
conv2 = conv2d(h_conv, w2, keep_prob) #conv
in_node = tf.nn.relu(conv2 + b2) #relu
up_h_convs[layer] = in_node
nConvFilter += 2
weights.append((w1, w2))
biases.append((b1, b2))
convs.append((conv1, conv2))
size = math.floor(size - (step * nLayerofConvolution))
with tf.name_scope('OUTPUT') as scope:
# Output Map
weight = weight_variable([1, 1, features_root, n_class], stddev)
bias = bias_variable([n_class])
conv = conv2d(in_node, weight, tf.constant(1.0)) #Last convolution 1x1
output_map = tf.nn.relu(conv + bias)
up_h_convs["out"] = output_map
nMaxPooling += 1
nConvFilter += 1
if summaries:
for i, (c1, c2) in enumerate(convs):
tf.summary.image('summary_conv_%02d_01' % i, get_image_summary(c1))
tf.summary.image('summary_conv_%02d_02' % i, get_image_summary(c2))
for k in pools.keys():
tf.summary.image('summary_pool_%02d' % k,
get_image_summary(pools[k]))
for k in deconv.keys():
tf.summary.image('summary_deconv_concat_%02d' % k,
get_image_summary(deconv[k]))
for k in dw_h_convs.keys():
tf.summary.histogram("dw_convolution_%02d" % k + '/activations',
dw_h_convs[k])
for k in up_h_convs.keys():
tf.summary.histogram("up_convolution_%s" % k + '/activations',
up_h_convs[k])
variables = []
for w1, w2 in weights:
variables.append(w1)
variables.append(w2)
for b1, b2 in biases:
variables.append(b1)
variables.append(b2)
print("\n\tpixelEdge = " + str(pixelEdge))
print("\tOffset pixels with input image = " + str(in_size - size) +
" pixels")
print("\tNumber of pixel to suppress : " + str((in_size - size) / 4) +
" by edge")
print('\tTotal Number of Convolution Filters : ' + str(nConvFilter))
print('\tNumber of max-pooling & deconvolution : ' + str(nMaxPooling))
return output_map, variables, int(in_size - size)