def __init__(self,
inplanes,
planes,
widefactor=1,
stride=1,
downsample=None,
num_domains=2):
super(Bottleneck, self).__init__()
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
self.bn1 = DomainSpecificBatchNorm2d(planes, num_domains)
self.conv2 = nn.Conv2d(planes,
planes * widefactor,
kernel_size=3,
stride=stride,
padding=1,
bias=False)
self.bn2 = DomainSpecificBatchNorm2d(planes * widefactor, num_domains)
self.conv3 = nn.Conv2d(planes * widefactor,
planes * 4,
kernel_size=1,
bias=False)
self.bn3 = DomainSpecificBatchNorm2d(planes * 4, num_domains)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
def __init__(self, inplanes, planes, stride=1, downsample=None, num_domains=2):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes, stride)
self.bn1 = DomainSpecificBatchNorm2d(planes, num_domains)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
self.bn2 = DomainSpecificBatchNorm2d(planes, num_domains)
self.downsample = downsample
self.stride = stride
def setup_net(self):
self.conv1 = nn.Conv2d(self.num_channels, 20, kernel_size=5)
self.bn1 = DomainSpecificBatchNorm2d(20, self.num_domains)
self.pool1 = nn.MaxPool2d(2)
self.conv2 = nn.Conv2d(20, 50, kernel_size=5)
self.bn2 = DomainSpecificBatchNorm2d(50, self.num_domains)
self.pool2 = nn.MaxPool2d(2)
self.fc1 = nn.Linear(50 * 4 * 4, 500)
if self.in_features != 0:
self.fc2 = nn.Linear(500, self.in_features)
self.fc3 = nn.Linear(self.in_features, self.num_classes)
else:
self.fc2 = nn.Linear(500, self.num_classes)
def _make_layer(self, block, planes, blocks, stride=1, num_domains=2):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = TwoInputSequential(
Conv2d(self.inplanes,
planes * block.expansion,
kernel_size=1,
stride=stride,
bias=False),
DomainSpecificBatchNorm2d(planes * block.expansion,
num_domains),
)
layers = []
layers.append(
block(self.inplanes,
planes,
stride,
downsample,
num_domains=num_domains))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(block(self.inplanes, planes,
num_domains=num_domains))
return TwoInputSequential(*layers)
def __init__(self, block, layers, in_features=256, num_classes=1000, num_domains=2):
self.inplanes = 64
self.in_features = in_features
self.num_domains = num_domains
self.num_classes = num_classes
super(DSBNResNet, self).__init__()
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = DomainSpecificBatchNorm2d(64, self.num_domains)
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], num_domains=self.num_domains)
self.layer2 = self._make_layer(block, 128, layers[1], stride=2, num_domains=self.num_domains)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2, num_domains=self.num_domains)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2, num_domains=self.num_domains)
# self.avgpool = nn.AvgPool2d(7, stride=1)
if self.in_features != 0:
self.fc1 = nn.Linear(512 * block.expansion, self.in_features)
self.fc2 = nn.Linear(self.in_features, num_classes)
else:
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
if isinstance(m, nn.Conv2d) or isinstance(m, Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.Linear):
m.weight.data.normal_(0, 0.01)
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()