def build(self, x, class_num, channel, wd, train=True):
self.conv1_1 = _conv_layer(x, [3, 3, channel, 64], 'conv1_1')
self.conv1_2 = _conv_layer(self.conv1_1, [3, 3, 64, 64], 'conv1_2')
self.pooling1 = _pooling_layer(self.conv1_2, 'pooling1')
self.conv2_1 = _conv_layer(self.pooling1, [3, 3, 64, 128], 'conv2_1')
self.conv2_2 = _conv_layer(self.conv2_1, [3, 3, 128, 128], 'conv2_2')
self.pooling2 = _pooling_layer(self.conv2_2, 'pooling2')
self.conv3_1 = _conv_layer(self.pooling2, [3, 3, 128, 256], 'conv3_1')
self.conv3_2 = _conv_layer(self.conv3_1, [3, 3, 256, 256], 'conv3_2')
self.conv3_3 = _conv_layer(self.conv3_2, [3, 3, 256, 256], 'conv3_3')
self.pooling3 = _pooling_layer(self.conv3_3, 'pooling3')
self.conv4_1 = _conv_layer(self.pooling3, [3, 3, 256, 512], 'conv4_1')
self.conv4_2 = _conv_layer(self.conv4_1, [3, 3, 512, 512], 'conv4_2')
self.conv4_3 = _conv_layer(self.conv4_2, [3, 3, 512, 512], 'conv4_3')
self.pooling4 = _pooling_layer(self.conv4_3, 'pooling4')
self.conv5_1 = _conv_layer(self.pooling4, [3, 3, 512, 512], 'conv5_1')
self.conv5_2 = _conv_layer(self.conv5_1, [3, 3, 512, 512], 'conv5_2')
self.conv5_3 = _conv_layer(self.conv5_2, [3, 3, 512, 512], 'conv5_3')
self.pooling5 = _pooling_layer(self.conv5_3, 'pooling5')
self.conv6 = _conv_layer(self.pooling5, [7, 7, 512, 4096], 'conv6')
if train:
self.conv6 = tf.nn.dropout(self.conv6, 0.5)
self.conv7 = _conv_layer(self.conv6, [1, 1, 4096, 4096], 'conv7')
if train:
self.conv7 = tf.nn.dropout(self.conv7, 0.5)
self.score_fr = _conv_layer_without_relu(self.conv7, [1, 1, 4096, 64],
'conv7_1x1conv',
wd=wd)
self.pooling4_conv = _conv_layer_without_relu(self.pooling4,
[1, 1, 512, 64],
'pool4_1x1conv',
wd=wd)
self.pooling3_conv = _conv_layer_without_relu(self.pooling3,
[1, 1, 256, 64],
'pool3_1x1conv',
wd=wd)
self.upconv5 = UpSampling2D((2, 2))(self.score_fr) #1/16
self.upconv4 = UpSampling2D((2, 2))(self.pooling4_conv) #1/8
self.upconv3 = UpSampling2D((8, 8))(self.pooling3_conv) #1
self.conv5_feature = self.Dense_decoder_feature_generation(
self.upconv5)
self.upconv4 = self.fusion1(self.conv5_feature, self.upconv4) #1/8
self.conv4_feature = self.Dense_decoder_feature_generation(
self.upconv4)
self.upconv3 = self.fusion2(self.conv5_feature, self.conv4_feature,
self.upconv3)
self.conv3_feature = self.Dense_decoder_feature_generation(
self.upconv3)
self.upscore = conv_bn_relu(
self.conv3_feature,
[3, 3, np.shape(self.conv3_feature)[3], 64], 1)
self.upscore = conv_bn_relu(self.upscore, [3, 3, 64, class_num], 1)
return self.upscore
def build(self, x, class_num, channel, wd, train=True):
self.conv1_1 = _conv_layer(x, [3, 3, channel, 64], 'conv1_1')
self.conv1_2 = _conv_layer(self.conv1_1, [3, 3, 64, 64], 'conv1_2')
self.pooling1 = _pooling_layer(self.conv1_2, 'pooling1')
self.conv2_1 = _conv_layer(self.pooling1, [3, 3, 64, 128], 'conv2_1')
self.conv2_2 = _conv_layer(self.conv2_1, [3, 3, 128, 128], 'conv2_2')
self.pooling2 = _pooling_layer(self.conv2_2, 'pooling2')
self.conv3_1 = _conv_layer(self.pooling2, [3, 3, 128, 256], 'conv3_1')
self.conv3_2 = _conv_layer(self.conv3_1, [3, 3, 256, 256], 'conv3_2')
self.conv3_3 = _conv_layer(self.conv3_2, [3, 3, 256, 256], 'conv3_3')
self.pooling3 = _pooling_layer(self.conv3_3, 'pooling3')
self.conv4_1 = _conv_layer(self.pooling3, [3, 3, 256, 512], 'conv4_1')
self.conv4_2 = _conv_layer(self.conv4_1, [3, 3, 512, 512], 'conv4_2')
self.conv4_3 = _conv_layer(self.conv4_2, [3, 3, 512, 512], 'conv4_3')
self.pooling4 = _pooling_layer(self.conv4_3, 'pooling4')
self.conv5_1 = _conv_layer(self.pooling4, [3, 3, 512, 512], 'conv5_1')
self.conv5_2 = _conv_layer(self.conv5_1, [3, 3, 512, 512], 'conv5_2')
self.conv5_3 = _conv_layer(self.conv5_2, [3, 3, 512, 512], 'conv5_3')
self.pooling5 = _pooling_layer(self.conv5_3, 'pooling5')
self.conv6 = _conv_layer(self.pooling5, [7, 7, 512, 4096], 'conv6')
if train:
self.conv6 = tf.nn.dropout(self.conv6, 0.5)
self.conv7 = _conv_layer(self.conv6, [1, 1, 4096, 4096], 'conv7')
if train:
self.conv7 = tf.nn.dropout(self.conv7, 0.5)
self.score_fr = _conv_layer_without_relu(self.conv7,
[1, 1, 4096, class_num],
'conv7_1x1conv',
wd=wd)
self.pooling4_conv = _conv_layer_without_relu(self.pooling4,
[1, 1, 512, class_num],
'pool4_1x1conv',
wd=wd)
self.pooling3_conv = _conv_layer_without_relu(self.pooling3,
[1, 1, 256, class_num],
'pool3_1x1conv',
wd=wd)
self.upscore = _upscore_layer(self.score_fr,
shape=tf.shape(self.pooling4_conv),
ksize=4,
stride=2,
num_class=class_num,
name='upscore')
self.fuse_pool4 = tf.add(self.pooling4_conv,
self.upscore,
name='fuse_pool4')
self.fuse_pool4_upscore = _upscore_layer(self.fuse_pool4,
shape=tf.shape(
self.pooling3_conv),
ksize=4,
stride=2,
num_class=class_num,
name='fuse_pool4_upscore')
self.fuse_pool3 = tf.add(self.pooling3_conv,
self.fuse_pool4_upscore,
name='fuse_pool3')
self.upscore = _upscore_layer(self.fuse_pool3,
shape=tf.shape(x),
ksize=16,
stride=8,
num_class=class_num,
name='final_upscore')
return self.upscore
def build(self, x, class_num, channel, wd, train=True):
# x: [batch_size,rows,cols,channel] batch_size = 256
self.conv1_1 = _conv_layer(x, [3, 3, channel, 64], 'conv1_1')
self.conv1_2 = _conv_layer(self.conv1_1, [3, 3, 64, 64], 'conv1_2')
self.pooling1 = _pooling_layer(self.conv1_2, 'pooling1')
self.conv2_1 = _conv_layer(self.pooling1, [3, 3, 64, 128], 'conv2_1')
self.conv2_2 = _conv_layer(self.conv2_1, [3, 3, 128, 128], 'conv2_2')
self.pooling2 = _pooling_layer(self.conv2_2, 'pooling2')
self.conv3_1 = _conv_layer(self.pooling2, [3, 3, 128, 256], 'conv3_1')
self.conv3_2 = _conv_layer(self.conv3_1, [3, 3, 256, 256], 'conv3_2')
self.conv3_3 = _conv_layer(self.conv3_2, [3, 3, 256, 256], 'conv3_3')
self.pooling3 = _pooling_layer(self.conv3_3, 'pooling3')
self.conv4_1 = _conv_layer(self.pooling3, [3, 3, 256, 512], 'conv4_1')
self.conv4_2 = _conv_layer(self.conv4_1, [3, 3, 512, 512], 'conv4_2')
self.conv4_3 = _conv_layer(self.conv4_2, [3, 3, 512, 512], 'conv4_3')
self.pooling4 = _pooling_layer(self.conv4_3, 'pooling4')
self.conv5_1 = _conv_layer(self.pooling4, [3, 3, 512, 512], 'conv5_1')
self.conv5_2 = _conv_layer(self.conv5_1, [3, 3, 512, 512], 'conv5_2')
self.conv5_3 = _conv_layer(self.conv5_2, [3, 3, 512, 512], 'conv5_3')
self.pooling5 = _pooling_layer(self.conv5_3, 'pooling5')
self.conv6 = _conv_layer(self.pooling5, [7, 7, 512, 4096], 'conv6')
if train:
self.conv6 = tf.nn.dropout(self.conv6, 0.5)
self.conv7 = _conv_layer(self.conv6, [1, 1, 4096, 4096], 'conv7')
if train:
self.conv7 = tf.nn.dropout(self.conv7, 0.5)
self.score_fr = _conv_layer_without_relu(self.conv7,
[1, 1, 4096, class_num],
'conv7_1x1conv',
wd=wd) #[256,4,8,2]
self.upscore = _upscore_layer(self.score_fr,
shape=tf.shape(x),
ksize=64,
stride=32,
num_class=class_num,
name='final_upscore')
return self.upscore
def build(self, x, class_num, channel, wd, train=True):
# x: [batch_size,rows,cols,channel]
# a = tf.expand_dims(x, 0)
# self.lstm1 = ConvLSTM2D(filters=channel,kernel_size=(1,1),padding='same',return_sequences=True)(a)
# i = tf.squeeze(self.lstm1,[0])
self.conv1_1 = _conv_layer(x, [3, 3, channel, 64], 'conv1_1')
self.conv1_2 = _conv_layer(self.conv1_1, [3, 3, 64, 64], 'conv1_2')
self.pooling1 = _pooling_layer(self.conv1_2, 'pooling1')
self.conv2_1 = _conv_layer(self.pooling1, [3, 3, 64, 128], 'conv2_1')
self.conv2_2 = _conv_layer(self.conv2_1, [3, 3, 128, 128], 'conv2_2')
self.pooling2 = _pooling_layer(self.conv2_2, 'pooling2')
self.conv3_1 = _conv_layer(self.pooling2, [3, 3, 128, 256], 'conv3_1')
self.conv3_2 = _conv_layer(self.conv3_1, [3, 3, 256, 256], 'conv3_2')
self.conv3_3 = _conv_layer(self.conv3_2, [3, 3, 256, 256], 'conv3_3')
self.pooling3 = _pooling_layer(self.conv3_3, 'pooling3')
self.conv4_1 = _conv_layer(self.pooling3, [3, 3, 256, 512], 'conv4_1')
self.conv4_2 = _conv_layer(self.conv4_1, [3, 3, 512, 512], 'conv4_2')
self.conv4_3 = _conv_layer(self.conv4_2, [3, 3, 512, 512], 'conv4_3')
self.pooling4 = _pooling_layer(self.conv4_3, 'pooling4')
self.conv5_1 = _conv_layer(self.pooling4, [3, 3, 512, 512], 'conv5_1')
self.conv5_2 = _conv_layer(self.conv5_1, [3, 3, 512, 512], 'conv5_2')
self.conv5_3 = _conv_layer(self.conv5_2, [3, 3, 512, 512], 'conv5_3')
self.pooling5 = _pooling_layer(self.conv5_3, 'pooling5')
self.conv6 = _conv_layer(self.pooling5, [7, 7, 512, 4096], 'conv6')
if train:
self.conv6 = tf.nn.dropout(self.conv6, 0.5)
self.conv7 = _conv_layer(self.conv6, [1, 1, 4096, 4096], 'conv7')
if train:
self.conv7 = tf.nn.dropout(self.conv7, 0.5)
self.score_fr = _conv_layer_without_relu(self.conv7,
[1, 1, 4096, class_num],
'conv7_1x1conv',
wd=wd)
#reshape score_fr ==> [(samples,time, rows, cols, channels)]
# self.score_fr = tf.expand_dims(self.score_fr, 0)
# self.lstm1 = ConvLSTM2D(filters=class_num,kernel_size=(1,1),padding='same',return_sequences=True)(self.score_fr)
self.pooling4_conv = _conv_layer_without_relu(self.pooling4,
[1, 1, 512, class_num],
'pool4_1x1conv',
wd=wd)
# reshape pooling4_conv ==> [(samples,time, rows, cols, channels)]
# self.pooling4_conv = tf.expand_dims(self.pooling4_conv, 0)
# self.lstm2 = ConvLSTM2D(filters=class_num, kernel_size=(1, 1), padding='same', return_sequences=True)(self.pooling4_conv)
self.pooling3_conv = _conv_layer_without_relu(self.pooling3,
[1, 1, 256, class_num],
'pool3_1x1conv',
wd=wd)
# reshape pooling3_conv ==> [(samples,time, rows, cols, channels)]
# self.pooling3_conv = tf.expand_dims(self.pooling3_conv, 0)
# self.lstm3 = ConvLSTM2D(filters=class_num, kernel_size=(1, 1), padding='same', return_sequences=True)(self.pooling3_conv)
#reshape lstm1,lstm2,lstm3 ,then fuse and upsample
# self.score_fr = tf.squeeze(self.lstm1, [0])
# self.pooling4_conv = tf.squeeze(self.lstm2, [0])
# self.pooling3_conv = tf.squeeze(self.lstm3, [0])
self.upscore = _upscore_layer(self.score_fr,
shape=tf.shape(self.pooling4_conv),
ksize=4,
stride=2,
num_class=class_num,
name='upscore')
self.fuse_pool4 = tf.add(self.pooling4_conv,
self.upscore,
name='fuse_pool4')
self.fuse_pool4_upscore = _upscore_layer(self.fuse_pool4,
shape=tf.shape(
self.pooling3_conv),
ksize=4,
stride=2,
num_class=class_num,
name='fuse_pool4_upscore')
self.fuse_pool3 = tf.add(self.pooling3_conv,
self.fuse_pool4_upscore,
name='fuse_pool3')
self.upscore = _upscore_layer(self.fuse_pool3,
shape=tf.shape(x),
ksize=16,
stride=8,
num_class=class_num,
name='final_upscore')
return self.upscore
def build(self, x, class_num, channel, wd, train=True):
#conv1
self.conv1_1 = _conv_layer(x, [3,3,channel, 64], 'conv1_1')
self.conv1_2 = _conv_layer(self.conv1_1, [3,3,64,64], 'conv1_2')
self.pooling1 = _pooling_layer(self.conv1_2, 'pooling1')
print("pooling1：", np.shape(self.pooling1))
self.conv2_1 = _conv_layer(self.pooling1, [3, 3, 64, 64], 'conv2_1')
self.conv2_2 = _conv_layer(self.conv2_1, [3, 3, 64, 64], 'conv2_2')
self.pooling2 = _pooling_layer(self.conv2_2, 'pooling2')
print("pooling2: ", np.shape(self.pooling2))
#conv2_x 3 blocks
self.conv2_block1 = bottleneck_block("conv2_block1", self.pooling2, 64, first_block=True, change_stride=False)
self.conv2_block2 = bottleneck_block("conv2_block2", self.conv2_block1, 64)
self.conv2_block3 = bottleneck_block("conv2_block3", self.conv2_block2, 64)
#conv3_x 4 blocks
self.conv3_block1 = bottleneck_block("conv3_block1", self.conv2_block3, 128, change_stride=True)
self.conv3_block2 = bottleneck_block("conv3_block2", self.conv3_block1, 128)
self.conv3_block3 = bottleneck_block("conv3_block3", self.conv3_block2, 128)
self.conv3_block4 = bottleneck_block("conv3_block4", self.conv3_block3, 128)
print("conv3_block4: ", np.shape(self.conv3_block4))
#conv4_x 23 blocks
self.conv4_block1 = bottleneck_block("conv4_block1", self.conv3_block4, 256, change_stride=True)
self.conv4_block2 = bottleneck_block("conv4_block2", self.conv4_block1, 256)
self.conv4_block3 = bottleneck_block("conv4_block3", self.conv4_block2, 256)
self.conv4_block4 = bottleneck_block("conv4_block4", self.conv4_block3, 256)
self.conv4_block5 = bottleneck_block("conv4_block5", self.conv4_block4, 256)
self.conv4_block6 = bottleneck_block("conv4_block6", self.conv4_block5, 256)
self.conv4_block7 = bottleneck_block("conv4_block7", self.conv4_block6, 256)
self.conv4_block8 = bottleneck_block("conv4_block8", self.conv4_block7, 256)
self.conv4_block9 = bottleneck_block("conv4_block9", self.conv4_block8, 256)
self.conv4_block10 = bottleneck_block("conv4_block10", self.conv4_block9, 256)
self.conv4_block11 = bottleneck_block("conv4_block11", self.conv4_block10, 256)
self.conv4_block12 = bottleneck_block("conv4_block12", self.conv4_block11, 256)
self.conv4_block13 = bottleneck_block("conv4_block13", self.conv4_block12, 256)
self.conv4_block14 = bottleneck_block("conv4_block14", self.conv4_block13, 256)
self.conv4_block15 = bottleneck_block("conv4_block15", self.conv4_block14, 256)
self.conv4_block16 = bottleneck_block("conv4_block16", self.conv4_block15, 256)
self.conv4_block17 = bottleneck_block("conv4_block17", self.conv4_block16, 256)
self.conv4_block18 = bottleneck_block("conv4_block18", self.conv4_block17, 256)
self.conv4_block19 = bottleneck_block("conv4_block19", self.conv4_block18, 256)
self.conv4_block20 = bottleneck_block("conv4_block20", self.conv4_block19, 256)
self.conv4_block21 = bottleneck_block("conv4_block21", self.conv4_block20, 256)
self.conv4_block22 = bottleneck_block("conv4_block22", self.conv4_block21, 256)
self.conv4_block23 = bottleneck_block("conv4_block23", self.conv4_block22, 256)
print(np.shape(self.conv4_block23))
#conv5_x 3 blocks
self.conv5_block1 = bottleneck_block("conv5_block1", self.conv4_block23, 512, change_stride=True)
self.conv5_block2 = bottleneck_block("conv5_block2", self.conv5_block1, 512)
self.conv5_block3 = bottleneck_block("conv5_block3", self.conv5_block2, 512)
print(np.shape(self.conv5_block3))
#1/32
self.score_32 = _conv_layer_without_relu(self.conv5_block3, [1,1,4*512,class_num], "1x1conv_1", wd=wd)
#1/16
self.score_16 = _conv_layer_without_relu(self.conv4_block23, [1,1,4*256,class_num], "1x1conv_2", wd=wd)
#1/8
self.score_8 = _conv_layer_without_relu(self.conv3_block4, [1,1,4*128,class_num], "1x1conv_3", wd=wd)
self.upscore1 = _upscore_layer(self.score_32, shape=tf.shape(self.score_16), ksize=4, stride=2, num_class=class_num, name="upscore1")
self.fuse1 = tf.add(self.score_16, self.upscore1, name="fuse1")
self.upscore2 = _upscore_layer(self.fuse1, shape=tf.shape(self.score_8), ksize=4, stride=2, num_class=class_num, name="upscore2")
self.fuse2 = tf.add(self.score_8, self.upscore2, name="fuse2")
self.upscore = _upscore_layer(self.fuse2, shape=tf.shape(x), ksize=16, stride=8, num_class=class_num, name="final_upscore")
return self.upscore
def build(self, x, class_num, channel, wd, train=True):
self.conv1_1 = _conv_layer(x, [3, 3, channel, 64], 'conv1_1')
self.conv1_2 = _conv_layer(self.conv1_1, [3, 3, 64, 64], 'conv1_2')
self.pooling1 = _pooling_layer(self.conv1_2, 'pooling1')
self.conv2_1 = _conv_layer(self.pooling1, [3, 3, 64, 128], 'conv2_1')
self.conv2_2 = _conv_layer(self.conv2_1, [3, 3, 128, 128], 'conv2_2')
self.pooling2 = _pooling_layer(self.conv2_2, 'pooling2')
# self.inception_block2_1 = _inception_layer(self.pooling1, in_c=64, out_1=32, out_21=32, out_22=32, out_31=32, out_32=64, out_33=64)
# self.inception_block2_2 = _inception_layer(self.inception_block2_1, in_c=128, out_1=32, out_21=32, out_22=32, out_31=32,
# out_32=64, out_33=64)
# self.pooling2 = _pooling_layer(self.inception_block2_2, 'pooling2')
# self.conv3_1 = _conv_layer(self.pooling2, [3, 3, 128, 256], 'conv3_1')
# self.conv3_2 = _conv_layer(self.conv3_1, [3, 3, 256, 256], 'conv3_2')
# self.conv3_3 = _conv_layer(self.conv3_2, [3, 3, 256, 256], 'conv3_3')
# self.pooling3 = _pooling_layer(self.conv3_3, 'pooling3')
self.inception_block3_1 = _inception_layer(
'inception3_1',
self.pooling2,
in_c=128,
out_1=64,
out_21=64,
out_22=64,
out_31=64,
out_32=128,
out_33=64,
out_41=64,
out_42=128,
out_43=64) # output channel 256
self.inception_block3_2 = _inception_layer(
'inception3_2',
self.inception_block3_1,
in_c=256,
out_1=64,
out_21=64,
out_22=64,
out_31=64,
out_32=128,
out_33=64,
out_41=64,
out_42=128,
out_43=64) # output channel 256
self.inception_block3_3 = _inception_layer(
'inception3_3',
self.inception_block3_2,
in_c=256,
out_1=64,
out_21=64,
out_22=64,
out_31=64,
out_32=128,
out_33=64,
out_41=64,
out_42=128,
out_43=64) # output channel 256
self.pooling3 = _pooling_layer(self.inception_block3_3, 'pooling3')
self.inception_block4_1 = _inception_layer('inception4_1',
self.pooling3,
in_c=256,
out_1=128,
out_21=128,
out_22=128,
out_31=128,
out_32=256,
out_33=128,
out_41=128,
out_42=256,
out_43=128) # output 512
self.inception_block4_2 = _inception_layer('inception4_2',
self.inception_block4_1,
in_c=512,
out_1=128,
out_21=128,
out_22=128,
out_31=128,
out_32=256,
out_33=128,
out_41=128,
out_42=256,
out_43=128) # output 512
self.inception_block4_3 = _inception_layer('inception4_3',
self.inception_block4_2,
in_c=512,
out_1=128,
out_21=128,
out_22=128,
out_31=128,
out_32=256,
out_33=128,
out_41=128,
out_42=256,
out_43=128) # output 512
self.pooling4 = _pooling_layer(self.inception_block4_3, 'pooling4')
self.inception_block5_1 = _inception_layer('inception5_1',
self.pooling4,
in_c=512,
out_1=128,
out_21=128,
out_22=128,
out_31=128,
out_32=256,
out_33=128,
out_41=128,
out_42=256,
out_43=128) # output 512
self.inception_block5_2 = _inception_layer('inception5_2',
self.inception_block5_1,
in_c=512,
out_1=128,
out_21=128,
out_22=128,
out_31=128,
out_32=256,
out_33=128,
out_41=128,
out_42=256,
out_43=128) # output 512
self.inception_block5_3 = _inception_layer('inception5_3',
self.inception_block5_2,
in_c=512,
out_1=128,
out_21=128,
out_22=128,
out_31=128,
out_32=256,
out_33=128,
out_41=128,
out_42=256,
out_43=128) # output 512
self.pooling5 = _pooling_layer(self.inception_block5_3, 'pooling5')
self.conv6 = _conv_layer(self.pooling5, [7, 7, 512, 4096], 'conv6')
if train:
self.conv6 = tf.nn.dropout(self.conv6, 0.5)
self.conv7 = _conv_layer(self.conv6, [1, 1, 4096, 4096], 'conv7')
if train:
self.conv7 = tf.nn.dropout(self.conv7, 0.5)
self.score_fr = _conv_layer_without_relu(self.conv7, [1, 1, 4096, 64],
'conv7_1x1conv',
wd=wd)
self.pooling4_conv = _conv_layer_without_relu(self.pooling4,
[1, 1, 512, 64],
'pool4_1x1conv',
wd=wd)
self.pooling3_conv = _conv_layer_without_relu(self.pooling3,
[1, 1, 256, 64],
'pool3_1x1conv',
wd=wd)
self.upconv5 = UpSampling2D((2, 2))(self.score_fr) #1/16
self.upconv4 = UpSampling2D((2, 2))(self.pooling4_conv) #1/8
self.upconv3 = UpSampling2D((8, 8))(self.pooling3_conv) #1
self.conv5_feature = self.Dense_decoder_feature_generation(
self.upconv5, name="gen1")
self.upconv4 = self.fusion1(self.conv5_feature, self.upconv4) #1/8
self.conv4_feature = self.Dense_decoder_feature_generation(
self.upconv4, name="gen2")
self.upconv3 = self.fusion2(self.conv5_feature, self.conv4_feature,
self.upconv3)
self.conv3_feature = self.Dense_decoder_feature_generation(
self.upconv3, name="gen3")
self.upscore = conv_bn_relu(
self.conv3_feature,
[3, 3, np.shape(self.conv3_feature)[3], 64],
1,
name="final1")
self.upscore = conv_bn_relu(self.upscore, [3, 3, 64, class_num],
1,
name="final2")
return self.upscore
def build(self, x, class_num, channel, wd, train=True):
#conv1
self.conv1_1 = _conv_layer(x, [3, 3, channel, 64], 'conv1_1')
self.conv1_2 = _conv_layer(self.conv1_1, [3, 3, 64, 64], 'conv1_2')
self.pooling1 = _pooling_layer(self.conv1_2, 'pooling1')
print("pooling1：", np.shape(self.pooling1))
self.conv2_1 = _conv_layer(self.pooling1, [3, 3, 64, 64], 'conv2_1')
self.conv2_2 = _conv_layer(self.conv2_1, [3, 3, 64, 64], 'conv2_2')
self.pooling2 = _pooling_layer(self.conv2_2, 'pooling2')
print("pooling2: ", np.shape(self.pooling2))
#conv2_x 3 blocks
self.conv2_block1 = residual_block("conv2_block1", self.pooling2, 64,
True)
self.conv2_block2 = residual_block("conv2_block2", self.conv2_block1,
64)
self.conv2_block3 = residual_block("conv2_block3", self.conv2_block2,
64)
#conv3_x 4 blocks
self.conv3_block1 = residual_block("conv3_block1", self.conv2_block3,
128)
self.conv3_block2 = residual_block("conv3_block2", self.conv3_block1,
128)
self.conv3_block3 = residual_block("conv3_block3", self.conv3_block2,
128)
self.conv3_block4 = residual_block("conv3_block4", self.conv3_block3,
128) #1/8
print("conv3_block4: ", np.shape(self.conv3_block4))
#conv4_x 6 blocks
self.conv4_block1 = residual_block("conv4_block1", self.conv3_block4,
256)
self.conv4_block2 = residual_block("conv4_block2", self.conv4_block1,
256)
self.conv4_block3 = residual_block("conv4_block3", self.conv4_block2,
256)
self.conv4_block4 = residual_block("conv4_block4", self.conv4_block3,
256)
self.conv4_block5 = residual_block("conv4_block5", self.conv4_block4,
256)
self.conv4_block6 = residual_block("conv4_block6", self.conv4_block5,
256) #1/16
print(np.shape(self.conv4_block6))
#conv5_x 3 blocks
self.conv5_block1 = residual_block("conv5_block1", self.conv4_block6,
512)
self.conv5_block2 = residual_block("conv5_block2", self.conv5_block1,
512)
self.conv5_block3 = residual_block("conv5_block3", self.conv5_block2,
512) #1/32
print(np.shape(self.conv5_block3))
#1/32
self.score_32 = _conv_layer_without_relu(self.conv5_block3,
[1, 1, 512, class_num],
"1x1conv_1",
wd=wd)
#1/16
self.score_16 = _conv_layer_without_relu(self.conv4_block6,
[1, 1, 256, class_num],
"1x1conv_2",
wd=wd)
#1/8
self.score_8 = _conv_layer_without_relu(self.conv3_block4,
[1, 1, 128, class_num],
"1x1conv_3",
wd=wd)
self.upscore1 = _upscore_layer(self.score_32,
shape=tf.shape(self.score_16),
ksize=4,
stride=2,
num_class=class_num,
name="upscore1")
self.fuse1 = tf.add(self.score_16, self.upscore1, name="fuse1")
self.upscore2 = _upscore_layer(self.fuse1,
shape=tf.shape(self.score_8),
ksize=4,
stride=2,
num_class=class_num,
name="upscore2")
self.fuse2 = tf.add(self.score_8, self.upscore2, name="fuse2")
self.upscore = _upscore_layer(self.fuse2,
shape=tf.shape(x),
ksize=16,
stride=8,
num_class=class_num,
name="final_upscore")
return self.upscore
