def make_stage(self, in_channel, out_channel, num, block):
bottleneck_channels = int(out_channel / 4)
down_block = PreActBottleNeck(in_channel, bottleneck_channels, 2)
self.add_block_list(down_block.get_name(), down_block, out_channel)
for _ in range(num):
temp_block = block(out_channel, out_channel)
self.add_block_list(temp_block.get_name(), temp_block, out_channel)
def create_block_list(self):
self.block_out_channels = []
self.index = 0
block1 = ConvBNActivationBlock(in_channels=self.data_channel,
out_channels=self.in_channels[0],
kernel_size=3,
padding=1,
bnName=self.bn_name,
activationName=self.activation_name)
self.add_block_list(block1.get_name(), block1, self.in_channels[0])
self.make_stage(self.in_channels[0], self.out_channels[0],
self.num_blocks[0], AttentionModule1)
self.make_stage(self.in_channels[1], self.out_channels[1],
self.num_blocks[1], AttentionModule2)
self.make_stage(self.in_channels[2], self.out_channels[2],
self.num_blocks[2], AttentionModule3)
output_channle = 2048
bottleneck_channels = int(output_channle / 4)
block2 = PreActBottleNeck(self.out_channels[2], bottleneck_channels, 2)
self.add_block_list(block2.get_name(), block2, output_channle)
block3 = PreActBottleNeck(output_channle, bottleneck_channels, 1)
self.add_block_list(block3.get_name(), block3, output_channle)
block4 = PreActBottleNeck(output_channle, bottleneck_channels, 1)
self.add_block_list(block4.get_name(), block4, output_channle)
def _make_residual(self, in_channels, out_channels, p):
layers = []
bottleneck_channels = int(out_channels / 4)
for _ in range(p):
layers.append(PreActBottleNeck(in_channels, bottleneck_channels, 1))
return nn.Sequential(*layers)
def __init__(self, in_channels, out_channels, p=1, t=2, r=1,
bn_name=NormalizationType.BatchNormalize2d,
activation_name=ActivationType.ReLU):
super().__init__(AttentionNetBlockName.AttentionModule3)
assert in_channels == out_channels
self.pre = self._make_residual(in_channels, out_channels, p)
self.trunk = self._make_residual(in_channels, out_channels, t)
self.soft_resdown1 = self._make_residual(in_channels, out_channels, r)
self.soft_resdown2 = self._make_residual(in_channels, out_channels, r)
self.soft_resup1 = self._make_residual(in_channels, out_channels, r)
self.soft_resup2 = self._make_residual(in_channels, out_channels, r)
bottleneck_channels = int(out_channels / 4)
self.shortcut = PreActBottleNeck(in_channels, bottleneck_channels, 1)
self.sigmoid = nn.Sequential(
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True),
nn.Conv2d(out_channels, out_channels, kernel_size=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True),
nn.Conv2d(out_channels, out_channels, kernel_size=1),
nn.Sigmoid()
)
self.last = self._make_residual(in_channels, out_channels, p)
def __init__(self, in_channels, out_channels, p=1, t=2, r=1,
bn_name=NormalizationType.BatchNormalize2d,
activation_name=ActivationType.ReLU):
super().__init__(AttentionNetBlockName.AttentionModule2)
# """The hyperparameter p denotes the number of preprocessing Residual
# Units before splitting into trunk branch and mask branch. t denotes
# the number of Residual Units in trunk branch. r denotes the number of
# Residual Units between adjacent pooling layer in the mask branch."""
assert in_channels == out_channels
self.pre = self._make_residual(in_channels, out_channels, p)
self.trunk = self._make_residual(in_channels, out_channels, t)
self.soft_resdown1 = self._make_residual(in_channels, out_channels, r)
self.soft_resdown2 = self._make_residual(in_channels, out_channels, r)
self.soft_resdown3 = self._make_residual(in_channels, out_channels, r)
self.soft_resup1 = self._make_residual(in_channels, out_channels, r)
self.soft_resup2 = self._make_residual(in_channels, out_channels, r)
self.soft_resup3 = self._make_residual(in_channels, out_channels, r)
bottleneck_channels = int(out_channels / 4)
self.shortcut = PreActBottleNeck(in_channels, bottleneck_channels, 1)
self.sigmoid = nn.Sequential(
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True),
nn.Conv2d(out_channels, out_channels, kernel_size=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True),
nn.Conv2d(out_channels, out_channels, kernel_size=1),
nn.Sigmoid()
)
self.last = self._make_residual(in_channels, out_channels, p)