def __init__(self, nclass, capacity=512, attention=False, drop=.4,
norm_layer=nn.BatchNorm, norm_kwargs=None, height=120, width=120):
super(_CAHead, self).__init__()
self.up_kwargs = {'height': height, 'width': width}
self.attention = attention
self.gamma = 1.0
height = height // 2
width = width // 2
with self.name_scope():
# Chained Context Aggregation Module
self.gp = GlobalFlow(capacity, 2048, norm_layer, norm_kwargs,
height=height, width=width)
self.cp1 = _ContextFlow(capacity, stride=2, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, height=height, width=width)
self.cp2 = _ContextFlow(capacity, stride=4, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, height=height, width=width)
self.cp3 = _ContextFlow(capacity, stride=8, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, height=height, width=width)
self.cp4 = _ContextFlow(capacity, stride=16, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, height=height, width=width)
if self.attention:
self.selection = _FeatureSelection(256, in_channels=capacity, norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
else:
self.proj = ConvBlock(256, 3, 1, 1, in_channels=capacity, norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.drop = nn.Dropout(drop) if drop else None
# decoder
self.decoder = ConvBlock(48, 3, 1, 1, norm_layer=norm_layer, norm_kwargs=norm_kwargs)
self.conv3x3 = ConvBlock(256, 3, 1, 1, norm_layer=norm_layer, norm_kwargs=norm_kwargs)
# segmentation head
self.seg_head = FCNHead(nclass, norm_layer=norm_layer, norm_kwargs=norm_kwargs)
def __init__(self, channels, in_channels, norm_layer=nn.BatchNorm, norm_kwargs=None):
super(_FeatureSelection, self).__init__()
with self.name_scope():
self.conv3x3 = ConvBlock(channels, 3, 1, 1, in_channels=in_channels,
norm_layer=norm_layer, norm_kwargs=norm_kwargs)
self.gap = nn.GlobalAvgPool2D()
self.conv1x1 = ConvBlock(channels, 1, in_channels=channels, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, activation='sigmoid')
def __init__(self, channels, norm_layer=nn.BatchNorm, norm_kwargs=None, reduction=1):
super(_FFModule, self).__init__()
with self.name_scope():
self.proj = ConvBlock(channels, 1, norm_layer=norm_layer, norm_kwargs=norm_kwargs)
self.gvp = nn.GlobalAvgPool2D()
self.conv1x1_1 = ConvBlock(channels // reduction, 1, norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.conv1x1_2 = ConvBlock(channels, 1, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, activation='sigmoid')
def __init__(self, channels, stride, groups=4, norm_layer=nn.BatchNorm,
norm_kwargs=None, height=60, width=60):
super(_ContextFlowShuffle, self).__init__()
self.stride = stride
self.groups = groups
self.up_kwargs = {'height': height, 'width': width}
with self.name_scope():
self.conv1 = ConvBlock(channels, 3, 1, 1, groups=groups, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, activation='relu')
self.conv2 = ConvBlock(channels, 3, 1, 1, groups=groups, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, activation='relu')
def __init__(self, channels, inter_channels=64, norm_layer=nn.BatchNorm,
norm_kwargs=None):
super(_SpatialPath, self).__init__()
with self.name_scope():
self.conv7x7 = ConvBlock(inter_channels, 7, 2, 3, norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.conv3x3_1 = ConvBlock(inter_channels, 3, 2, 1, norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.conv3x3_2 = ConvBlock(inter_channels, 3, 2, 1, norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.conv1x1 = ConvBlock(channels, 1, norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
def __init__(self, nclass, low_channels=256, high_channels=128,
norm_layer=nn.BatchNorm, norm_kwargs=None, drop=.1):
super(_BoundaryAttention, self).__init__()
with self.name_scope():
self.conv1x1 = ConvBlock(low_channels, 1, in_channels=high_channels, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, activation='sigmoid')
self.fconv1x1 = ConvBlock(high_channels, 1, in_channels=low_channels,
norm_layer=norm_layer, norm_kwargs=norm_kwargs)
self.fconv3x3 = ConvBlock(high_channels, 3, 1, 1, in_channels=high_channels,
norm_layer=norm_layer, norm_kwargs=norm_kwargs)
self.cconv3x3 = ConvBlock(high_channels, 3, 1, 1, in_channels=high_channels,
norm_layer=norm_layer, norm_kwargs=norm_kwargs)
self.drop = nn.Dropout(drop) if drop else None
self.cconv1x1 = nn.Conv2D(nclass, 1, in_channels=high_channels)
def __init__(self,
nclass,
decoder_capacity,
input_height,
input_width,
norm_layer=nn.BatchNorm,
norm_kwargs=None):
super(_LadderHead, self).__init__()
with self.name_scope():
self.conv_c4 = ConvBlock(decoder_capacity,
1,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation='relu')
self.fusion_c3 = _LadderFusion(decoder_capacity,
input_height // 16,
input_width // 16,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.fusion_c2 = _LadderFusion(decoder_capacity,
input_height // 8,
input_width // 8,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.fusion_c1 = _LadderFusion(decoder_capacity,
input_height // 4,
input_width // 4,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.seg_head = FCNHead(nclass,
decoder_capacity,
norm_layer,
norm_kwargs,
activation='relu')
def __init__(self,
in_channels,
channels,
kernel_size,
strides=1,
padding=0,
dilation=1,
norm_layer=nn.BatchNorm,
norm_kwargs=None,
activation='relu'):
super(SeparableConvBlock, self).__init__()
self.conv = nn.Conv2D(in_channels,
kernel_size,
strides,
padding,
dilation,
groups=in_channels,
use_bias=False,
in_channels=in_channels)
self.pointwise = ConvBlock(channels,
1,
use_bias=False,
in_channels=in_channels,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation=activation)
def __init__(self,
capacity,
height,
width,
norm_layer=nn.BatchNorm,
norm_kwargs=None):
super(_LadderFusion, self).__init__()
with self.name_scope():
self.conv1x1 = ConvBlock(capacity,
1,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation='relu')
self.conv3x3 = ConvBlock(capacity,
3,
1,
1,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation='relu')
self.up_kwargs = {'height': height, 'width': width}
def __init__(self,
in_channels,
mid_channels,
strides,
dilation=1,
norm_layer=nn.BatchNorm,
norm_kwargs=None,
activation='relu'):
super(Block, self).__init__()
if strides > 1:
self.down = ConvBlock(mid_channels * self.expansion,
1,
strides=strides,
use_bias=False,
in_channels=in_channels,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation=None)
else:
self.down = None
self.residual = nn.HybridSequential()
self.residual.add(
SeparableConvBlock(in_channels,
mid_channels,
3,
strides,
dilation,
dilation,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation=activation),
SeparableConvBlock(mid_channels,
mid_channels,
3,
1,
1,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation=activation),
SeparableConvBlock(mid_channels,
mid_channels * self.expansion,
3,
1,
1,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation=None))
self.act = Activation(activation)
def __init__(self, nclass, height, width, norm_layer=nn.BatchNorm, norm_kwargs=None):
super(_BiSeNetHead, self).__init__()
self.up_kwargs = {'height': height, 'width': width}
with self.name_scope():
self.spatial_path = _SpatialPath(128, norm_layer=norm_layer, norm_kwargs=norm_kwargs)
self.global_flow = GlobalFlow(128, in_channels=512, norm_layer=norm_layer,
norm_kwargs=norm_kwargs, height=height, width=width)
self.refine_c4 = _ARModule(128, in_channels=512, norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.refine_c3 = _ARModule(128, in_channels=256, norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.proj = ConvBlock(128, 3, 1, 1, norm_layer=norm_layer, norm_kwargs=norm_kwargs)
self.fusion = _FFModule(256, norm_layer, norm_kwargs, reduction=1)
self.seg = FCNHead(nclass, in_channels=256, norm_layer=norm_layer, norm_kwargs=norm_kwargs,
drop_out=.0, reduction=4)
def __init__(self,
nclass,
in_channels,
norm_layer=nn.BatchNorm,
norm_kwargs=None,
height=60,
width=60):
super(_DeepLabHead, self).__init__()
self.up_kwargs = {'height': height, 'width': width}
with self.name_scope():
self.aspp = ASPP(256,
in_channels,
norm_layer,
norm_kwargs,
height // 2,
width // 2,
atrous_rates=(12, 24, 36))
self.conv_c1 = ConvBlock(48,
3,
1,
1,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.conv3x3 = ConvBlock(256,
3,
1,
1,
in_channels=304,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.drop = nn.Dropout(0.5)
self.head = FCNHead(nclass,
256,
norm_layer,
norm_kwargs,
reduction=1)
def __init__(self,
nclass,
num_scale=2,
height=60,
width=60,
norm_layer=nn.BatchNorm,
norm_kwargs=None):
super(_AttentionHead, self).__init__()
self.num_scale = num_scale
self.up_kwargs = {'height': height, 'width': width}
with self.name_scope():
self.seg_head = FCNHead(nclass,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.conv3x3 = ConvBlock(512,
3,
1,
1,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation='relu')
self.conv1x1 = nn.Conv2D(num_scale, 1, in_channels=512)
def __init__(self,
block,
layers,
channels,
classes=1000,
norm_layer=nn.BatchNorm,
norm_kwargs=None,
**kwargs):
super(Xception, self).__init__()
self.in_channels = 8
self.norm_kwargs = norm_kwargs if norm_kwargs is not None else {}
with self.name_scope():
self.conv1 = ConvBlock(self.in_channels,
3,
2,
1,
use_bias=False,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs,
activation='relu')
self.maxpool = nn.MaxPool2D(pool_size=3, strides=2, padding=1)
self.layer1 = self._make_layer(block,
norm_layer,
layers[0],
channels[0],
strides=2)
self.layer2 = self._make_layer(block,
norm_layer,
layers[1],
channels[1],
strides=2)
self.layer3 = self._make_layer(block,
norm_layer,
layers[2],
channels[2],
strides=2)
self.avgpool = nn.GlobalAvgPool2D()
self.flat = nn.Flatten()
self.fc = nn.Dense(in_units=self.in_channels, units=classes)
def __init__(self,
nclass,
input_height,
input_width,
capacity=128,
norm_layer=nn.BatchNorm,
norm_kwargs=None):
super(_SwiftNetHead, self).__init__()
with self.name_scope():
self.conv1x1 = ConvBlock(capacity,
1,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.fusion_32x = _LateralFusion(capacity,
input_height // 32,
input_width // 32,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.fusion_16x = _LateralFusion(capacity,
input_height // 16,
input_width // 16,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.fusion_8x = _LateralFusion(capacity,
input_height // 8,
input_width // 8,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.final = _LateralFusion(capacity,
input_height // 4,
input_width // 4,
True,
norm_layer=norm_layer,
norm_kwargs=norm_kwargs)
self.seg_head = FCNHead(nclass,
capacity,
norm_layer,
norm_kwargs,
reduction=1)