def __init__(self, inplanes, planes, baseWidth, cardinality, stride=1, downsample=None):
""" Constructor
Args:
inplanes: input channel dimensionality
planes: output channel dimensionality
baseWidth: base width.
cardinality: num of convolution groups.
stride: conv stride. Replaces pooling layer.
"""
super(Bottleneck, self).__init__()
D = int(math.floor(planes * (baseWidth / 64)))
C = cardinality
self.conv1 = L.Conv2d(inplanes, D * C, kernel_size=1, stride=1, padding=0, bias=False)
self.bn1 = L.BatchNorm2d(D * C)
self.conv2 = L.Conv2d(
D * C, D * C, kernel_size=3, stride=stride, padding=1, groups=C, bias=False
)
self.bn2 = L.BatchNorm2d(D * C)
self.conv3 = L.Conv2d(D * C, planes * 4, kernel_size=1, stride=1, padding=0, bias=False)
self.bn3 = L.BatchNorm2d(planes * 4)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
def __init__(
self,
in_planes,
out_planes,
kernel_size,
stride=1,
padding=0,
dilation=1,
groups=1,
bias=False,
):
super(BasicConv2d, self).__init__()
self.basicconv = nn.Sequential(
L.Conv2d(
in_planes,
out_planes,
kernel_size=kernel_size,
stride=stride,
padding=padding,
dilation=dilation,
groups=groups,
bias=bias,
),
L.BatchNorm2d(out_planes),
nn.ReLU(inplace=True),
)
def _make_layer(self, block, planes, blocks, stride=1):
"""Stack n bottleneck modules where n is inferred from the depth of the network.
Args:
block: block type used to construct ResNext
planes: number of output channels (need to multiply by block.expansion)
blocks: number of blocks to be built
stride: factor to reduce the spatial dimensionality in the first bottleneck of the block.
Returns: a Module consisting of n sequential bottlenecks.
"""
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
L.Conv2d(
self.inplanes,
planes * block.expansion,
kernel_size=1,
stride=stride,
bias=False,
),
L.BatchNorm2d(planes * block.expansion),
)
layers = []
layers.append(
block(self.inplanes, planes, self.baseWidth, self.cardinality,
stride, downsample))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(
block(self.inplanes, planes, self.baseWidth, self.cardinality))
return nn.Sequential(*layers)
def __init__(self, baseWidth, cardinality, layers, num_classes):
"""Constructor
Args:
baseWidth: baseWidth for ResNeXt.
cardinality: number of convolution groups.
layers: config of layers, e.g., [3, 4, 6, 3]
num_classes: number of classes
"""
super(ResNeXt, self).__init__()
block = Bottleneck
self.cardinality = cardinality
self.baseWidth = baseWidth
self.num_classes = num_classes
self.inplanes = 64
self.output_size = 64
self.conv1 = L.Conv2d(3, 64, 7, 2, 3, bias=False)
self.bn1 = L.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.maxpool1 = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], 2)
self.layer3 = self._make_layer(block, 256, layers[2], 2)
self.layer4 = self._make_layer(block, 512, layers[3], 2)
self.avgpool = nn.AvgPool2d(7)
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2.0 / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
def __init__(self, block, layers, num_classes=1000, zero_init_residual=False):
super(ResNet, self).__init__()
self.inplanes = 64
self.conv1 = L.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
self.bn1 = L.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
# Zero-initialize the last BN in each residual branch,
# so that the residual branch starts with zeros, and each residual block behaves like an identity.
# This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
if zero_init_residual:
for m in self.modules():
if isinstance(m, Bottleneck):
nn.init.constant_(m.bn3.weight, 0)
elif isinstance(m, BasicBlock):
nn.init.constant_(m.bn2.weight, 0)
def conv1x1(in_planes, out_planes, stride=1):
"""1x1 convolution"""
return L.Conv2d(in_planes,
out_planes,
kernel_size=1,
stride=stride,
bias=False)
def conv3x3(in_planes, out_planes, stride=1):
"""3x3 convolution with padding"""
return L.Conv2d(in_planes,
out_planes,
kernel_size=3,
stride=stride,
padding=1,
bias=False)
def Backbone_WSGNR50_Custumed(in_C):
net = l_resnet50(pretrained=True)
if in_C != 3:
net.conv1 = L.Conv2d(in_C,
64,
kernel_size=7,
stride=2,
padding=3,
bias=False)
div_2 = nn.Sequential(*list(net.children())[:3])
div_4 = nn.Sequential(*list(net.children())[3:5])
div_8 = net.layer2
div_16 = net.layer3
div_32 = net.layer4
return div_2, div_4, div_8, div_16, div_32
def __init__(self, h_C, l_C):
super(SIM, self).__init__()
self.h2l_pool = nn.AvgPool2d((2, 2), stride=2)
self.l2h_up = cus_sample
self.h2l_0 = L.Conv2d(h_C, l_C, 3, 1, 1)
self.h2h_0 = L.Conv2d(h_C, h_C, 3, 1, 1)
self.bnl_0 = L.BatchNorm2d(l_C)
self.bnh_0 = L.BatchNorm2d(h_C)
self.h2h_1 = L.Conv2d(h_C, h_C, 3, 1, 1)
self.h2l_1 = L.Conv2d(h_C, l_C, 3, 1, 1)
self.l2h_1 = L.Conv2d(l_C, h_C, 3, 1, 1)
self.l2l_1 = L.Conv2d(l_C, l_C, 3, 1, 1)
self.bnl_1 = L.BatchNorm2d(l_C)
self.bnh_1 = L.BatchNorm2d(h_C)
self.h2h_2 = L.Conv2d(h_C, h_C, 3, 1, 1)
self.l2h_2 = L.Conv2d(l_C, h_C, 3, 1, 1)
self.bnh_2 = L.BatchNorm2d(h_C)
self.relu = nn.ReLU(True)
def __init__(self, in_hc=64, in_lc=256, out_c=64, main=0):
super(conv_2nV1, self).__init__()
self.main = main
mid_c = min(in_hc, in_lc)
self.relu = nn.ReLU(True)
self.h2l_pool = nn.AvgPool2d((2, 2), stride=2)
self.l2h_up = nn.Upsample(scale_factor=2, mode="nearest")
# stage 0
self.h2h_0 = L.Conv2d(in_hc, mid_c, 3, 1, 1)
self.l2l_0 = L.Conv2d(in_lc, mid_c, 3, 1, 1)
self.bnh_0 = L.BatchNorm2d(mid_c)
self.bnl_0 = L.BatchNorm2d(mid_c)
# stage 1
self.h2h_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.h2l_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.l2h_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.l2l_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.bnl_1 = L.BatchNorm2d(mid_c)
self.bnh_1 = L.BatchNorm2d(mid_c)
if self.main == 0:
# stage 2
self.h2h_2 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.l2h_2 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.bnh_2 = L.BatchNorm2d(mid_c)
# stage 3
self.h2h_3 = L.Conv2d(mid_c, out_c, 3, 1, 1)
self.bnh_3 = L.BatchNorm2d(out_c)
self.identity = L.Conv2d(in_hc, out_c, 1)
elif self.main == 1:
# stage 2
self.h2l_2 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.l2l_2 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.bnl_2 = L.BatchNorm2d(mid_c)
# stage 3
self.l2l_3 = L.Conv2d(mid_c, out_c, 3, 1, 1)
self.bnl_3 = L.BatchNorm2d(out_c)
self.identity = L.Conv2d(in_lc, out_c, 1)
else:
raise NotImplementedError
def __init__(self, in_hc=64, in_mc=256, in_lc=512, out_c=64):
super(conv_3nV1, self).__init__()
self.upsample = nn.Upsample(scale_factor=2, mode="nearest")
self.downsample = nn.AvgPool2d((2, 2), stride=2)
mid_c = 64
self.relu = nn.ReLU(True)
# stage 0
self.h2h_0 = L.Conv2d(in_hc, mid_c, 3, 1, 1)
self.m2m_0 = L.Conv2d(in_mc, mid_c, 3, 1, 1)
self.l2l_0 = L.Conv2d(in_lc, mid_c, 3, 1, 1)
self.bnh_0 = L.BatchNorm2d(mid_c)
self.bnm_0 = L.BatchNorm2d(mid_c)
self.bnl_0 = L.BatchNorm2d(mid_c)
# stage 1
self.h2h_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.h2m_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.m2h_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.m2m_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.m2l_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.l2m_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.l2l_1 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.bnh_1 = L.BatchNorm2d(mid_c)
self.bnm_1 = L.BatchNorm2d(mid_c)
self.bnl_1 = L.BatchNorm2d(mid_c)
# stage 2
self.h2m_2 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.l2m_2 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.m2m_2 = L.Conv2d(mid_c, mid_c, 3, 1, 1)
self.bnm_2 = L.BatchNorm2d(mid_c)
# stage 3
self.m2m_3 = L.Conv2d(mid_c, out_c, 3, 1, 1)
self.bnm_3 = L.BatchNorm2d(out_c)
self.identity = L.Conv2d(in_mc, out_c, 1)
