Method __init__

features/resnet_features.py:130–170

(self, block, layers, num_classes=1000, zero_init_residual=False)

    '''

    def __init__(self, block, layers, num_classes=1000, zero_init_residual=False):
        super(ResNet_features, self).__init__()

        self.inplanes = 64

        # the first convolutional layer before the structured sequence of blocks
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
                               bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        # comes from the first conv and the following max pool
        self.kernel_sizes = [7, 3]
        self.strides = [2, 2]
        self.paddings = [3, 1]

        # the following layers, each layer is a sequence of blocks
        self.block = block
        self.layers = layers
        self.layer1 = self._make_layer(block=block, planes=64, num_blocks=self.layers[0])
        self.layer2 = self._make_layer(block=block, planes=128, num_blocks=self.layers[1], stride=2)
        self.layer3 = self._make_layer(block=block, planes=256, num_blocks=self.layers[2], stride=2)
        self.layer4 = self._make_layer(block=block, planes=512, num_blocks=self.layers[3], stride=2)

        # initialize the parameters
        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 _make_layer(self, block, planes, num_blocks, stride=1):
        downsample = None