def __init__(self, in_channels, n_classes):
super(VoxResNet, self).__init__()
self.conv1 = nn.Conv3d(in_channels=in_channels, out_channels=32, kernel_size=3, padding=1)
self.bn1 = nn.BatchNorm3d(32)
self.act1 = ActFunc('ReLU')
self.conv2 = nn.Conv3d(in_channels=32, out_channels=32, kernel_size=3, padding=1)
self.bn2 = nn.BatchNorm3d(32)
self.act2 = ActFunc('ReLU')
self.conv3 = nn.Conv3d(in_channels=32, out_channels=64, kernel_size=3, padding=1, stride=2)
self.mod1 = VoxResModule(in_channels=64)
self.mod2 = VoxResModule(in_channels=64)
self.bn3 = nn.BatchNorm3d(64)
self.act3 = ActFunc('ReLU')
self.conv4 = nn.Conv3d(in_channels=64, out_channels=64, kernel_size=3, padding=1, stride=2)
self.mod3 = VoxResModule(in_channels=64)
self.mod4 = VoxResModule(in_channels=64)
self.bn4 = nn.BatchNorm3d(64)
self.act4 = ActFunc('ReLU')
self.conv5 = nn.Conv3d(in_channels=64, out_channels=64, kernel_size=3, padding=1, stride=2)
self.mod5 = VoxResModule(in_channels=64)
self.mod6 = VoxResModule(in_channels=64)
# Deconvolution layers
self.deconv1 = nn.ConvTranspose3d(in_channels=32, out_channels=n_classes, kernel_size=3, stride=1, padding=1)
self.deconv2 = nn.ConvTranspose3d(in_channels=64, out_channels=n_classes, kernel_size=2, stride=2)
self.deconv3 = nn.ConvTranspose3d(in_channels=64, out_channels=n_classes, kernel_size=4, stride=4)
self.deconv4 = nn.ConvTranspose3d(in_channels=64, out_channels=n_classes, kernel_size=8, stride=8)
self.softmax = F.softmax
def __init__(self, in_channels):
super(VoxResModule, self).__init__()
self.bn1 = nn.BatchNorm3d(in_channels)
self.act1 = ActFunc('ReLU')
self.conv1 = nn.Conv3d(in_channels=in_channels, out_channels=in_channels, kernel_size=3, padding=1)
self.bn2 = nn.BatchNorm3d(in_channels)
self.act2 = ActFunc('ReLU')
self.conv2 = nn.Conv3d(in_channels=in_channels, out_channels=in_channels, kernel_size=3, padding=1)
def __init__(self, in_channels, n_convs):
super(DownTransition, self).__init__()
out_channels = 2 * in_channels
self.down_conv = nn.Conv3d(in_channels=in_channels, out_channels=out_channels, kernel_size=2, stride=2, bias=False)
self.bn1 = nn.BatchNorm3d(out_channels)
self.act1 = ActFunc('PReLU', num_parameters=out_channels)
self.conv_block = _make_conv_layer(out_channels, n_convs)
self.act2 = ActFunc('PReLU', num_parameters=out_channels)
def __init__(self, in_channels, out_channels, n_convs):
super(UpTransition, self).__init__()
self.de_conv = nn.ConvTranspose3d(in_channels=in_channels,
out_channels=out_channels // 2,
kernel_size=2,
stride=2,
bias=False)
self.bn1 = nn.BatchNorm3d(out_channels // 2)
self.act1 = ActFunc('PReLU', num_parameters=out_channels // 2)
self.dropout1 = CutoutDropout(p=0.2)
self.conv_block = _make_conv_layer(out_channels, n_convs)
self.act2 = ActFunc('PReLU', num_parameters=out_channels)
self.dropout2 = CutoutDropout(p=0.2)
def __init__(self, in_channels, gating_channels, inter_channels):
super(AttentionGate, self).__init__()
self.wx = nn.Conv3d(in_channels=in_channels,
out_channels=inter_channels,
kernel_size=1,
stride=1,
bias=False)
self.wg = nn.Conv3d(in_channels=gating_channels,
out_channels=inter_channels,
kernel_size=1,
stride=1,
bias=False)
self.psi = nn.Conv3d(in_channels=inter_channels,
out_channels=1,
kernel_size=1,
stride=1,
bias=False)
self.act1 = ActFunc('ReLU')
self.act2 = nn.Sigmoid()
self.wout = nn.Sequential(
nn.Conv3d(in_channels=in_channels,
out_channels=in_channels,
kernel_size=1,
stride=1,
bias=False), nn.BatchNorm3d(in_channels))
def __init__(self, in_channels):
super(OutputTransition, self).__init__()
self.conv1 = nn.Conv3d(in_channels=in_channels, out_channels=2, kernel_size=5, padding=2, bias=False)
self.bn1 = nn.BatchNorm3d(2)
self.conv2 = nn.Conv3d(in_channels=2, out_channels=2, kernel_size=1, bias=False)
self.act1 = ActFunc('PReLU', num_parameters=2)
self.softmax = F.softmax
def __init__(self):
super(InputTransition, self).__init__()
self.conv1 = nn.Conv3d(in_channels=1,
out_channels=16,
kernel_size=5,
padding=2,
bias=False)
self.bn1 = nn.BatchNorm3d(16)
self.act1 = ActFunc('PReLU', num_parameters=16)
def __init__(self, in_channels, out_channels, kernel_size=1):
super(GatingSignal, self).__init__()
self.conv = nn.Conv3d(in_channels=in_channels,
out_channels=out_channels,
kernel_size=kernel_size,
stride=1,
padding=0,
bias=False)
self.bn = nn.BatchNorm3d(out_channels)
self.act = ActFunc("ReLU")
def _make_conv_layer(in_channels, n_convs):
layers = []
for i in range(n_convs):
# Adding non-linearity
if i != 0:
layers.append(ActFunc('PReLU', num_parameters=in_channels))
layers.append(ConvBlock(in_channels=in_channels, out_channels=in_channels))
return nn.Sequential(*layers)
def __init__(self, in_channels, out_channels):
super(MPRBlock, self).__init__()
assert out_channels % 4 == 0
inter_channels = out_channels // 4
self.axial = nn.Conv3d(in_channels=in_channels,
out_channels=inter_channels,
kernel_size=(1, 3, 3),
stride=(1, 1, 1),
padding=(0, 1, 1),
bias=False)
self.bn_axial = nn.BatchNorm3d(inter_channels)
self.act_axial = ActFunc('ReLU')
self.coronal = nn.Conv3d(in_channels=in_channels,
out_channels=inter_channels,
kernel_size=(3, 1, 3),
stride=(1, 1, 1),
padding=(1, 0, 1),
bias=False)
self.bn_coronal = nn.BatchNorm3d(inter_channels)
self.act_coronal = ActFunc('ReLU')
self.sagittal = nn.Conv3d(in_channels=in_channels,
out_channels=inter_channels,
kernel_size=(3, 3, 1),
stride=(1, 1, 1),
padding=(1, 1, 0),
bias=False)
self.bn_sagittal = nn.BatchNorm3d(inter_channels)
self.act_sagittal = ActFunc('ReLU')
self.all = nn.Conv3d(in_channels=in_channels,
out_channels=inter_channels,
kernel_size=3,
stride=1,
padding=1,
bias=False)
self.bn_all = nn.BatchNorm3d(inter_channels)
self.act_all = ActFunc('ReLU')
def _make_layer(self, block, in_channels, growth_rate, nb_layers, drop_rate):
layers = []
layers.append(nn.Conv3d(in_channels=in_channels,
out_channels=growth_rate,
kernel_size=3,
stride=1,
padding=1))
layers.append(nn.BatchNorm3d(growth_rate))
layers.append(ActFunc('ReLU'))
for i in range(nb_layers-1):
layers.append(block((i+1)*growth_rate, growth_rate, drop_rate))
return nn.Sequential(*layers)