def __init__(self):
super(Generator,self).__init__()
self.fc1 = Linear(input_dim=64,output_dim=1024)
self.bn1 = BatchNorm(num_channels=256,act='relu')
self.conv2 = Conv2DTranspose(num_channels=256,num_filters=192,filter_size=3,padding=1,output_size=4,stride=2)
self.bn2 = BatchNorm(num_channels=192,act='relu')
self.conv3 = Conv2DTranspose(num_channels=192,num_filters=128,filter_size=3,padding=1,output_size=8,stride=2)
self.bn3 = BatchNorm(num_channels=128,act='relu')
self.conv4 = Conv2DTranspose(num_channels=128,num_filters=64,filter_size=3,padding=1,output_size=16,stride=2)
self.bn4 = BatchNorm(num_channels=64,act='relu')
self.conv5 = Conv2DTranspose(num_channels=64,num_filters=1,filter_size=3,padding=1,output_size=32,stride=2,act='tanh')
def __init__(self, num_channels, num_filters):
super(Decoder, self).__init__()
# TODO: decoder contains:
# 1 2x2 transpose conv (makes feature map 2x larger)
# 1 3x3 conv + 1bn + 1relu +
# 1 3x3 conv + 1bn + 1relu
self.up = Conv2DTranspose(num_channels=num_channels,
num_filters=num_filters,
filter_size=2,
stride=2)
self.conv1 = Conv2D(num_channels,
num_filters,
filter_size=3,
stride=1,
padding=1)
self.bn1 = BatchNorm(num_filters, act='relu')
self.conv2 = Conv2D(num_filters,
num_filters,
filter_size=3,
stride=1,
padding=1)
self.bn2 = BatchNorm(num_filters, act='relu')
def __init__(self, num_classes=59):
super(FCN8s, self).__init__()
backbone = VGG16BN(pretrained=False)
self.layer1 = backbone.layer1
self.layer1[0].conv._padding = [100, 100]
self.pool1 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.layer2 = backbone.layer2
self.pool2 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.layer3 = backbone.layer3
self.pool3 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.layer4 = backbone.layer4
self.pool4 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.layer5 = backbone.layer5
self.pool5 = Pool2D(pool_size=2, pool_stride=2, ceil_mode=True)
self.fc6 = Conv2D(512, 4096, 7, act='relu')
self.fc7 = Conv2D(4096, 4096, 1, act='relu')
self.drop6 = Dropout()
self.drop7 = Dropout()
self.score = Conv2D(4096, num_classes, 1)
self.score_pool3 = Conv2D(256, num_classes, 1)
self.score_pool4 = Conv2D(512, num_classes, 1)
self.up_output = Conv2DTranspose(num_channels=num_classes,
num_filters=num_classes,
filter_size=4,
stride=2,
bias_attr=False)
self.up_pool4 = Conv2DTranspose(num_channels=num_classes,
num_filters=num_classes,
filter_size=4,
stride=2,
bias_attr=False)
self.up_final = Conv2DTranspose(num_channels=num_classes,
num_filters=num_classes,
filter_size=16,
stride=8,
bias_attr=False)
def __init__(self,num_classes=59):
super(FCN8s,self).__init__()
vgg16bn = VGG16BN()
self.layer1 = vgg16bn.layer1
self.layer1[0].conv._padding = [100,100]
self.layer2 = vgg16bn.layer2
self.layer3 = vgg16bn.layer3
self.layer4 = vgg16bn.layer4
self.layer5 = vgg16bn.layer5
# self.conv1_1 = Conv2D(3,64,3,padding=1)
# self.conv1_2 = Conv2D(64,64,3,padding=1)
self.pool1 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
# self.conv2_1 = Conv2D(64,128,3,padding=1)
# self.conv2_2 = Conv2D(128,128,3,padding=1)
self.pool2 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
# self.conv3_1 = Conv2D(128,256,3,padding=1)
# self.conv3_2 = Conv2D(256,256,3,padding=1)
# self.conv3_3 = Conv2D(256,256,3,padding=1)
self.pool3 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
# self.conv4_1 = Conv2D(256,512,3,padding=1)
# self.conv4_2 = Conv2D(512,512,3,padding=1)
# self.conv4_3 = Conv2D(512,512,3,padding=1)
self.pool4 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
# self.conv5_1 = Conv2D(512,512,3,padding=1)
# self.conv5_2 = Conv2D(512,512,3,padding=1)
# self.conv5_3 = Conv2D(512,512,3,padding=1)
self.pool5 = Pool2D(pool_size=2,pool_stride=2,ceil_mode=True)
self.conv6 = Conv2D(512,4096,1,act='relu')
self.conv7 = Conv2D(4096,4096,1,act='relu')
self.drop6 = Dropout()
self.drop7 = Dropout()
self.score = Conv2D(4096,num_classes,1)
self.score_pool3 = Conv2D(256,num_classes,1)
self.score_pool4 = Conv2D(512,num_classes,1,)
self.upsample1 = Conv2DTranspose(num_classes,num_classes,filter_size=4,stride=2,padding=2,bias_attr=False)
self.upsample2 = Conv2DTranspose(num_classes,num_classes,filter_size=4,stride=2,padding=2,bias_attr=False)
self.upsample3 = Conv2DTranspose(num_classes,num_classes,filter_size=16,stride=8,padding=1,bias_attr=False)
def __init__(self, name_scope, md=4):
super(PWCDCNet, self).__init__(name_scope)
self.param_attr = fluid.ParamAttr(
name='conv_weights',
regularizer=fluid.regularizer.L2DecayRegularizer(
regularization_coeff=0.0004),
initializer=fluid.initializer.MSRAInitializer(uniform=True,
fan_in=None,
seed=0))
self.md = md
self.conv1a = Conv2D("conv1a",
16,
filter_size=3,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv1aa = Conv2D("conv1aa",
16,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv1b = Conv2D("conv1b",
16,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv2a = Conv2D("conv2a",
32,
filter_size=3,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv2aa = Conv2D("conv2aa",
32,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv2b = Conv2D("conv2b",
32,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv3a = Conv2D("conv3a",
64,
filter_size=3,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv3aa = Conv2D("conv3aa",
64,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv3b = Conv2D("conv3b",
64,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv4a = Conv2D("conv4a",
96,
filter_size=3,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv4aa = Conv2D("conv4aa",
96,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv4b = Conv2D("conv4b",
96,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv5a = Conv2D("conv5a",
128,
filter_size=3,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv5aa = Conv2D("conv5aa",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv5b = Conv2D("conv5b",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv6aa = Conv2D("conv6aa",
196,
filter_size=3,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv6a = Conv2D("conv6a",
196,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv6b = Conv2D("conv6b",
196,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv6_0 = Conv2D("conv6_0",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv6_1 = Conv2D("conv6_1",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv6_2 = Conv2D("conv6_2",
96,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv6_3 = Conv2D("conv6_3",
64,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv6_4 = Conv2D("conv6_4",
32,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.predict_flow6 = Conv2D("predict_flow6",
2,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.deconv6 = Conv2DTranspose("deconv6",
2,
filter_size=4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.upfeat6 = Conv2DTranspose("upfeat6",
2,
filter_size=4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv5_0 = Conv2D("conv5_0",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv5_1 = Conv2D("conv5_1",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv5_2 = Conv2D("conv5_2",
96,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv5_3 = Conv2D("conv5_3",
64,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv5_4 = Conv2D("conv5_4",
32,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.predict_flow5 = Conv2D("predict_flow5",
2,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.deconv5 = Conv2DTranspose("deconv5",
2,
filter_size=4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.upfeat5 = Conv2DTranspose("upfeat5",
2,
filter_size=4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv4_0 = Conv2D("conv4_0",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv4_1 = Conv2D("conv4_1",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv4_2 = Conv2D("conv4_2",
96,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv4_3 = Conv2D("conv4_3",
64,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv4_4 = Conv2D("conv4_4",
32,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.predict_flow4 = Conv2D("predict_flow4",
2,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.deconv4 = Conv2DTranspose("deconv4",
2,
filter_size=4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.upfeat4 = Conv2DTranspose("upfeat4",
2,
filter_size=4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv3_0 = Conv2D("conv3_0",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv3_1 = Conv2D("conv3_1",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv3_2 = Conv2D("conv3_2",
96,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv3_3 = Conv2D("conv3_3",
64,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv3_4 = Conv2D("conv3_4",
32,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.predict_flow3 = Conv2D("predict_flow3",
2,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.deconv3 = Conv2DTranspose("deconv3",
2,
filter_size=4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.upfeat3 = Conv2DTranspose("upfeat3",
2,
filter_size=4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.conv2_0 = Conv2D("conv2_0",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv2_1 = Conv2D("conv2_1",
128,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv2_2 = Conv2D("conv2_2",
96,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv2_3 = Conv2D("conv2_3",
64,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.conv2_4 = Conv2D("conv2_4",
32,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.predict_flow2 = Conv2D("predict_flow2",
2,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
self.deconv2 = Conv2DTranspose("deconv2",
2,
filter_size=4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.dc_conv1 = Conv2D("dc_conv1",
128,
filter_size=3,
stride=1,
padding=1,
dilation=1,
param_attr=self.param_attr)
self.dc_conv2 = Conv2D("dc_conv2",
128,
filter_size=3,
stride=1,
padding=2,
dilation=2,
param_attr=self.param_attr)
self.dc_conv3 = Conv2D("dc_conv3",
128,
filter_size=3,
stride=1,
padding=4,
dilation=4,
param_attr=self.param_attr)
self.dc_conv4 = Conv2D("dc_conv4",
96,
filter_size=3,
stride=1,
padding=8,
dilation=8,
param_attr=self.param_attr)
self.dc_conv5 = Conv2D("dc_conv5",
64,
filter_size=3,
stride=1,
padding=16,
dilation=16,
param_attr=self.param_attr)
self.dc_conv6 = Conv2D("dc_conv6",
32,
filter_size=3,
stride=1,
padding=1,
dilation=1,
param_attr=self.param_attr)
self.dc_conv7 = Conv2D("dc_conv7",
2,
filter_size=3,
stride=1,
padding=1,
param_attr=self.param_attr)
def __init__(self, input_channels=1):
'''
:功能 :类初始化函数
:参数 input_channels :int,输入数据的通道数，1或3或4
:返回 :None
'''
super(GGCNN, self).__init__()
#网络结构定义，直接照搬GGCNN 三层卷积三层反卷积
self.conv1 = Conv2D(input_channels,
filter_sizes[0],
kernel_sizes[0],
stride=strides[0],
padding=3,
act='relu',
param_attr=fluid.initializer.Xavier(uniform=False))
self.conv2 = Conv2D(filter_sizes[0],
filter_sizes[1],
kernel_sizes[1],
stride=strides[1],
padding=2,
act='relu',
param_attr=fluid.initializer.Xavier(uniform=False))
self.conv3 = Conv2D(filter_sizes[1],
filter_sizes[2],
kernel_sizes[2],
stride=strides[2],
padding=1,
act='relu',
param_attr=fluid.initializer.Xavier(uniform=False))
self.convt1 = Conv2DTranspose(
filter_sizes[2],
filter_sizes[3],
kernel_sizes[3],
stride=strides[3],
act='relu',
output_size=50,
padding=1,
param_attr=fluid.initializer.Xavier(
uniform=False)) #这里本来有个output_padding参数，但paddle没有，故删了
self.convt2 = Conv2DTranspose(
filter_sizes[3],
filter_sizes[4],
kernel_sizes[4],
stride=strides[4],
act='relu',
output_size=100,
padding=2,
param_attr=fluid.initializer.Xavier(uniform=False))
self.convt3 = Conv2DTranspose(
filter_sizes[4],
filter_sizes[5],
kernel_sizes[5],
stride=strides[5],
act='relu',
output_size=301,
padding=3,
param_attr=fluid.initializer.Xavier(uniform=False))
self.pos_output = Conv2D(
filter_sizes[5],
1,
filter_size=2,
param_attr=fluid.initializer.Xavier(uniform=False))
self.cos_output = Conv2D(
filter_sizes[5],
1,
filter_size=2,
param_attr=fluid.initializer.Xavier(uniform=False))
self.sin_output = Conv2D(
filter_sizes[5],
1,
filter_size=2,
param_attr=fluid.initializer.Xavier(uniform=False))
self.width_output = Conv2D(
filter_sizes[5],
1,
filter_size=2,
param_attr=fluid.initializer.Xavier(uniform=False))
def __init__(self, md=4):
super(PWCDCNet, self).__init__()
self.md = md
self.param_attr = fluid.ParamAttr(
regularizer=fluid.regularizer.L2DecayRegularizer(
regularization_coeff=0.0004),
initializer=fluid.initializer.MSRAInitializer(uniform=True,
fan_in=None,
seed=0))
self.conv1a = Conv2D(3, 16, 3, 2, 1, param_attr=self.param_attr)
self.conv1aa = Conv2D(16, 16, 3, 1, 1, param_attr=self.param_attr)
self.conv1b = Conv2D(16, 16, 3, 1, 1, param_attr=self.param_attr)
self.conv2a = Conv2D(16, 32, 3, 2, 1, param_attr=self.param_attr)
self.conv2aa = Conv2D(32, 32, 3, 1, 1, param_attr=self.param_attr)
self.conv2b = Conv2D(32, 32, 3, 1, 1, param_attr=self.param_attr)
self.conv3a = Conv2D(32, 64, 3, 2, 1, param_attr=self.param_attr)
self.conv3aa = Conv2D(64, 64, 3, 1, 1, param_attr=self.param_attr)
self.conv3b = Conv2D(64, 64, 3, 1, 1, param_attr=self.param_attr)
self.conv4a = Conv2D(64, 96, 3, 2, 1, param_attr=self.param_attr)
self.conv4aa = Conv2D(96, 96, 3, 1, 1, param_attr=self.param_attr)
self.conv4b = Conv2D(96, 96, 3, 1, 1, param_attr=self.param_attr)
self.conv5a = Conv2D(96, 128, 3, 2, 1, param_attr=self.param_attr)
self.conv5aa = Conv2D(128, 128, 3, 1, 1, param_attr=self.param_attr)
self.conv5b = Conv2D(128, 128, 3, 1, 1, param_attr=self.param_attr)
self.conv6aa = Conv2D(128, 196, 3, 2, 1, param_attr=self.param_attr)
self.conv6a = Conv2D(196, 196, 3, 1, 1, param_attr=self.param_attr)
self.conv6b = Conv2D(196, 196, 3, 1, 1, param_attr=self.param_attr)
nd = (2 * self.md + 1)**2
dd = np.cumsum([128, 128, 96, 64, 32], dtype=np.int32).astype(np.int)
dd = [int(d) for d in dd]
od = nd
self.conv6_0 = Conv2D(od, 128, 3, 1, 1, param_attr=self.param_attr)
self.conv6_1 = Conv2D(od + dd[0],
128,
3,
1,
1,
param_attr=self.param_attr)
self.conv6_2 = Conv2D(od + dd[1],
96,
3,
1,
1,
param_attr=self.param_attr)
self.conv6_3 = Conv2D(od + dd[2],
64,
3,
1,
1,
param_attr=self.param_attr)
self.conv6_4 = Conv2D(od + dd[3],
32,
3,
1,
1,
param_attr=self.param_attr)
self.predict_flow6 = Conv2D(od + dd[4],
2,
3,
1,
1,
param_attr=self.param_attr)
self.deconv6 = Conv2DTranspose(2,
2,
4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.upfeat6 = Conv2DTranspose(od + dd[4],
2,
4,
stride=2,
padding=1,
param_attr=self.param_attr)
od = nd + 128 + 4
self.conv5_0 = Conv2D(od, 128, 3, 1, 1, param_attr=self.param_attr)
self.conv5_1 = Conv2D(od + dd[0],
128,
3,
1,
1,
param_attr=self.param_attr)
self.conv5_2 = Conv2D(od + dd[1],
96,
3,
1,
1,
param_attr=self.param_attr)
self.conv5_3 = Conv2D(od + dd[2],
64,
3,
1,
1,
param_attr=self.param_attr)
self.conv5_4 = Conv2D(od + dd[3],
32,
3,
1,
1,
param_attr=self.param_attr)
self.predict_flow5 = Conv2D(od + dd[4],
2,
3,
1,
1,
param_attr=self.param_attr)
self.deconv5 = Conv2DTranspose(2,
2,
4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.upfeat5 = Conv2DTranspose(od + dd[4],
2,
4,
stride=2,
padding=1,
param_attr=self.param_attr)
od = nd + 96 + 4
self.conv4_0 = Conv2D(od, 128, 3, 1, 1, param_attr=self.param_attr)
self.conv4_1 = Conv2D(od + dd[0],
128,
3,
1,
1,
param_attr=self.param_attr)
self.conv4_2 = Conv2D(od + dd[1],
96,
3,
1,
1,
param_attr=self.param_attr)
self.conv4_3 = Conv2D(od + dd[2],
64,
3,
1,
1,
param_attr=self.param_attr)
self.conv4_4 = Conv2D(od + dd[3],
32,
3,
1,
1,
param_attr=self.param_attr)
self.predict_flow4 = Conv2D(od + dd[4],
2,
3,
1,
1,
param_attr=self.param_attr)
self.deconv4 = Conv2DTranspose(2,
2,
4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.upfeat4 = Conv2DTranspose(od + dd[4],
2,
4,
stride=2,
padding=1,
param_attr=self.param_attr)
od = nd + 64 + 4
self.conv3_0 = Conv2D(od, 128, 3, 1, 1, param_attr=self.param_attr)
self.conv3_1 = Conv2D(od + dd[0],
128,
3,
1,
1,
param_attr=self.param_attr)
self.conv3_2 = Conv2D(od + dd[1],
96,
3,
1,
1,
param_attr=self.param_attr)
self.conv3_3 = Conv2D(od + dd[2],
64,
3,
1,
1,
param_attr=self.param_attr)
self.conv3_4 = Conv2D(od + dd[3],
32,
3,
1,
1,
param_attr=self.param_attr)
self.predict_flow3 = Conv2D(od + dd[4],
2,
3,
1,
1,
param_attr=self.param_attr)
self.deconv3 = Conv2DTranspose(2,
2,
4,
stride=2,
padding=1,
param_attr=self.param_attr)
self.upfeat3 = Conv2DTranspose(od + dd[4],
2,
4,
stride=2,
padding=1,
param_attr=self.param_attr)
od = nd + 32 + 4
self.conv2_0 = Conv2D(od, 128, 3, 1, 1, param_attr=self.param_attr)
self.conv2_1 = Conv2D(od + dd[0],
128,
3,
1,
1,
param_attr=self.param_attr)
self.conv2_2 = Conv2D(od + dd[1],
96,
3,
1,
1,
param_attr=self.param_attr)
self.conv2_3 = Conv2D(od + dd[2],
64,
3,
1,
1,
param_attr=self.param_attr)
self.conv2_4 = Conv2D(od + dd[3],
32,
3,
1,
1,
param_attr=self.param_attr)
self.predict_flow2 = Conv2D(od + dd[4],
2,
3,
1,
1,
param_attr=self.param_attr)
# self.deconv2 = Conv2DTranspose(2, 2, 4, stride=2, padding=1, param_attr=self.param_attr)
self.dc_conv1 = Conv2D(od + dd[4],
128,
3,
1,
1,
dilation=1,
param_attr=self.param_attr)
self.dc_conv2 = Conv2D(128,
128,
3,
1,
2,
dilation=2,
param_attr=self.param_attr)
self.dc_conv3 = Conv2D(128,
128,
3,
1,
4,
dilation=4,
param_attr=self.param_attr)
self.dc_conv4 = Conv2D(128,
96,
3,
1,
8,
dilation=8,
param_attr=self.param_attr)
self.dc_conv5 = Conv2D(96,
64,
3,
1,
16,
dilation=16,
param_attr=self.param_attr)
self.dc_conv6 = Conv2D(64,
32,
3,
1,
1,
dilation=1,
param_attr=self.param_attr)
self.dc_conv7 = Conv2D(32, 2, 3, 1, 1, param_attr=self.param_attr)