def attention(scale_input, is_training=False):
l2_reg = FLAGS.learning_rate
dropout_ratio = 0
if is_training is True:
dropout_ratio = 0.5
conv1 = Utils.conv(scale_input, filters=512, l2_reg_scale=l2_reg)
conv1 = Utils.dropout(conv1, dropout_ratio, is_training)
conv2 = Utils.conv(conv1, filters=3, kernel_size=[
1, 1], l2_reg_scale=l2_reg)
return conv2
def attention(scale_input, is_training=False):
l2_reg = FLAGS.learning_rate
keep_prob = 1.0
if is_training is True:
keep_prob = 0.5
with tf.variable_scope("attention"):
conv1 = Utils.conv(scale_input, filters=512, l2_reg_scale=l2_reg)
conv1 = Utils.dropout(conv1, keep_prob)
conv2 = Utils.conv(conv1,
filters=3,
kernel_size=[1, 1],
l2_reg_scale=l2_reg)
return conv2
def attention_gate(encoder_input, decoder_input, filters, is_training=False):
l2_reg = FLAGS.learning_rate
decoder_input = tf.image.resize_images(decoder_input,
tf.shape(encoder_input)[1:3, ])
decoder_input = Utils.conv(decoder_input,
filters=filters,
l2_reg_scale=l2_reg,
is_training=is_training)
gated = tf.reduce_mean(tf.stack([encoder_input, decoder_input]), axis=0)
gated = tf.nn.relu(gated)
gated = Utils.conv(gated,
filters=filters,
l2_reg_scale=l2_reg,
is_training=is_training)
gated = tf.nn.sigmoid(gated)
gated = tf.multiply(gated, encoder_input)
return gated
def unetinference(image, keep_prob):
net = {}
l2_reg = FLAGS.learning_rate
# added for resume better
global_iter_counter = tf.Variable(0, name='global_step', trainable=False)
net['global_step'] = global_iter_counter
with tf.variable_scope("inference"):
inputs = image
teacher = tf.placeholder(
tf.float32, [None, IMAGE_SIZE, IMAGE_SIZE, NUM_OF_CLASSES])
is_training = True
# 1, 1, 3
conv1_1 = Utils.conv(inputs,
filters=64,
l2_reg_scale=l2_reg,
batchnorm_istraining=is_training)
conv1_2 = Utils.conv(conv1_1,
filters=64,
l2_reg_scale=l2_reg,
batchnorm_istraining=is_training)
pool1 = Utils.pool(conv1_2)
# 1/2, 1/2, 64
conv2_1 = Utils.conv(pool1,
filters=128,
l2_reg_scale=l2_reg,
batchnorm_istraining=is_training)
conv2_2 = Utils.conv(conv2_1,
filters=128,
l2_reg_scale=l2_reg,
batchnorm_istraining=is_training)
pool2 = Utils.pool(conv2_2)
# 1/4, 1/4, 128
conv3_1 = Utils.conv(pool2,
filters=256,
l2_reg_scale=l2_reg,
batchnorm_istraining=is_training)
conv3_2 = Utils.conv(conv3_1,
filters=256,
l2_reg_scale=l2_reg,
batchnorm_istraining=is_training)
pool3 = Utils.pool(conv3_2)
# 1/8, 1/8, 256
conv4_1 = Utils.conv(pool3,
filters=512,
l2_reg_scale=l2_reg,
batchnorm_istraining=is_training)
conv4_2 = Utils.conv(conv4_1,
filters=512,
l2_reg_scale=l2_reg,
batchnorm_istraining=is_training)
pool4 = Utils.pool(conv4_2)
# 1/16, 1/16, 512
conv5_1 = Utils.conv(pool4, filters=1024, l2_reg_scale=l2_reg)
conv5_2 = Utils.conv(conv5_1, filters=1024, l2_reg_scale=l2_reg)
concated1 = tf.concat([
Utils.conv_transpose(conv5_2, filters=512, l2_reg_scale=l2_reg),
conv4_2
],
axis=3)
conv_up1_1 = Utils.conv(concated1, filters=512, l2_reg_scale=l2_reg)
conv_up1_2 = Utils.conv(conv_up1_1, filters=512, l2_reg_scale=l2_reg)
concated2 = tf.concat([
Utils.conv_transpose(conv_up1_2, filters=256, l2_reg_scale=l2_reg),
conv3_2
],
axis=3)
conv_up2_1 = Utils.conv(concated2, filters=256, l2_reg_scale=l2_reg)
conv_up2_2 = Utils.conv(conv_up2_1, filters=256, l2_reg_scale=l2_reg)
concated3 = tf.concat([
Utils.conv_transpose(conv_up2_2, filters=128, l2_reg_scale=l2_reg),
conv2_2
],
axis=3)
conv_up3_1 = Utils.conv(concated3, filters=128, l2_reg_scale=l2_reg)
conv_up3_2 = Utils.conv(conv_up3_1, filters=128, l2_reg_scale=l2_reg)
concated4 = tf.concat([
Utils.conv_transpose(conv_up3_2, filters=64, l2_reg_scale=l2_reg),
conv1_2
],
axis=3)
conv_up4_1 = Utils.conv(concated4, filters=64, l2_reg_scale=l2_reg)
conv_up4_2 = Utils.conv(conv_up4_1, filters=64, l2_reg_scale=l2_reg)
outputs = Utils.conv(conv_up4_2,
filters=NUM_OF_CLASSES,
kernel_size=[1, 1],
activation=None)
annotation_pred = tf.argmax(outputs, dimension=3, name="prediction")
return tf.expand_dims(annotation_pred, dim=3), outputs, net, conv5_2
def u_net_inference(image, is_training=False):
net = {}
l2_reg = FLAGS.learning_rate
# added for resume better
global_iter_counter = tf.Variable(0, name='global_step', trainable=False)
net['global_step'] = global_iter_counter
with tf.variable_scope("inference"):
inputs = image
# 1, 1, 3 Encoder 1st
conv1_1 = Utils.conv(inputs,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training)
conv1_2 = Utils.conv(conv1_1,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training)
pool1 = Utils.pool(conv1_2)
# 1/2, 1/2, 64 Encoder 2nd
conv2_1 = Utils.conv(pool1,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training)
conv2_2 = Utils.conv(conv2_1,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training)
pool2 = Utils.pool(conv2_2)
# 1/4, 1/4, 128 Encoder 3rd
conv3_1 = Utils.conv(pool2,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training)
conv3_2 = Utils.conv(conv3_1,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training)
pool3 = Utils.pool(conv3_2)
# 1/8, 1/8, 256 Encoder 4th
conv4_1 = Utils.conv(pool3,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training)
conv4_2 = Utils.conv(conv4_1,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training)
pool4 = Utils.pool(conv4_2)
# 1/16, 1/16, 512 Encoder 5th, upsample, skip 1
conv5_1 = Utils.conv(pool4,
filters=1024,
l2_reg_scale=l2_reg,
is_training=is_training)
conv5_2 = Utils.conv(conv5_1,
filters=1024,
l2_reg_scale=l2_reg,
is_training=is_training)
concated1 = tf.concat([
Utils.conv_transpose(conv5_2,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training), conv4_2
],
axis=3)
# Decoder 1st, skip 2
conv_up1_1 = Utils.conv(concated1,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training)
conv_up1_2 = Utils.conv(conv_up1_1,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training)
concated2 = tf.concat([
Utils.conv_transpose(conv_up1_2,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training), conv3_2
],
axis=3)
# Decoder 2nd, skip 3
conv_up2_1 = Utils.conv(concated2,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training)
conv_up2_2 = Utils.conv(conv_up2_1,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training)
concated3 = tf.concat([
Utils.conv_transpose(conv_up2_2,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training), conv2_2
],
axis=3)
# Decoder 3rd, skip 4
conv_up3_1 = Utils.conv(concated3,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training)
conv_up3_2 = Utils.conv(conv_up3_1,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training)
concated4 = tf.concat([
Utils.conv_transpose(conv_up3_2,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training), conv1_2
],
axis=3)
# Decoder 4th
conv_up4_1 = Utils.conv(concated4,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training)
conv_up4_2 = Utils.conv(conv_up4_1,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training)
# logits/probability
logits = Utils.conv(conv_up4_2,
filters=NUM_OF_CLASSES,
kernel_size=[1, 1],
activation=None)
# Output/prediction
annotation_pred = tf.argmax(logits, dimension=3, name="prediction")
outputs = tf.expand_dims(annotation_pred, dim=3)
return outputs, logits, net, conv5_2
def unetinference(image, is_training=False):
net = {}
l2_reg = FLAGS.learning_rate
# added for resume better
global_iter_counter = tf.Variable(0, name='global_step', trainable=False)
net['global_step'] = global_iter_counter
with tf.variable_scope("inference"):
inputs = image
# 1, 1, 3
conv1_1 = Utils.conv(inputs,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training)
conv1_2 = Utils.conv(conv1_1,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training)
pool1 = Utils.pool(conv1_2)
# 1/2, 1/2, 64
conv2_1 = Utils.conv(pool1,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training)
conv2_2 = Utils.conv(conv2_1,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training)
pool2 = Utils.pool(conv2_2)
# 1/4, 1/4, 128
conv3_1 = Utils.conv(pool2,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training)
conv3_2 = Utils.conv(conv3_1,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training)
pool3 = Utils.pool(conv3_2)
# 1/8, 1/8, 256
conv4_1 = Utils.conv(pool3,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training)
conv4_2 = Utils.conv(conv4_1,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training)
pool4 = Utils.pool(conv4_2)
# 1/16, 1/16, 512
conv5_1 = Utils.conv(pool4,
filters=1024,
l2_reg_scale=l2_reg,
is_training=is_training)
conv5_2 = Utils.conv(conv5_1,
filters=1024,
l2_reg_scale=l2_reg,
is_training=is_training)
concated1 = tf.concat([
Utils.conv_transpose(conv5_2,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training),
attention_gate(conv4_2, conv5_2, 512)
],
axis=3)
conv_up1_1 = Utils.conv(concated1,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training)
conv_up1_2 = Utils.conv(conv_up1_1,
filters=512,
l2_reg_scale=l2_reg,
is_training=is_training)
concated2 = tf.concat([
Utils.conv_transpose(conv_up1_2,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training),
attention_gate(conv3_2, conv_up1_2, 256)
],
axis=3)
conv_up2_1 = Utils.conv(concated2,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training)
conv_up2_2 = Utils.conv(conv_up2_1,
filters=256,
l2_reg_scale=l2_reg,
is_training=is_training)
concated3 = tf.concat([
Utils.conv_transpose(conv_up2_2,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training),
attention_gate(conv2_2, conv_up2_2, 128)
],
axis=3)
conv_up3_1 = Utils.conv(concated3,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training)
conv_up3_2 = Utils.conv(conv_up3_1,
filters=128,
l2_reg_scale=l2_reg,
is_training=is_training)
concated4 = tf.concat([
Utils.conv_transpose(conv_up3_2,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training),
attention_gate(conv1_2, conv_up3_2, 64)
],
axis=3)
conv_up4_1 = Utils.conv(concated4,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training)
conv_up4_2 = Utils.conv(conv_up4_1,
filters=64,
l2_reg_scale=l2_reg,
is_training=is_training)
logits = Utils.conv(conv_up4_2,
filters=NUM_OF_CLASSES,
kernel_size=[1, 1],
activation=None,
is_training=is_training)
annotation_pred = tf.argmax(logits, dimension=3, name="prediction")
return tf.expand_dims(annotation_pred, dim=3), logits, net
