October 3, 2019

nn.Module: Network Construction

Background

To define our network, we should succeed class nn.Module and implement the function forward. We put all the layers we want in the function __init__() and define how layers connect in function forward.

Example

class Hopenet(nn.Module):
    # Hopenet with 3 output layers for yaw, pitch and roll
    # Predicts Euler angles by binning and regression with the expected value
    def __init__(self, block, layers, num_bins):
        self.inplanes = 64
        super(Hopenet, self).__init__()
        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)
        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.AvgPool2d(7)
        self.fc_yaw = nn.Linear(512 * block.expansion, num_bins)
        self.fc_pitch = nn.Linear(512 * block.expansion, num_bins)
        self.fc_roll = nn.Linear(512 * block.expansion, num_bins)

        # Vestigial layer from previous experiments
        self.fc_finetune = nn.Linear(512 * block.expansion + 3, 3)

        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. / n))
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()

1.Constructor

super(Hopenet, self).__init__(): call the construtor in superclass

2.Layers

  • nn.Conv2d:
    class torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True)
  • nn.BatchNorm2d:
    class torch.nn.BatchNorm2d(num_features, eps=1e-05, momentum=0.1, affine=True)
    For each channel (the second dimension), the batch normalization computes the mean and variance once.
  • nn.ReLU:
    class torch.nn.ReLU(inplace=False)
    The inplace will cover the old value with a new value and save the memory.
  • nn.MaxPool2d:
    class torch.nn.MaxPool2d(kernel_size, stride=None, padding=0, dilation=1, return_indices=False, ceil_mode=False)
  • nn.AvgPool2d:
    class torch.nn.AvgPool2d(kernel_size, stride=None, padding=0, ceil_mode=False, count_include_pad=True)
  • nn.Linear:
    class torch.nn.Linear(in_features, out_features, bias=True)

3.Initialization

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. / n))
    elif isinstance(m, nn.BatchNorm2d):
        m.weight.data.fill_(1)
        m.bias.data.zero_()

The kernel is the weight and the gamma and beta are the weight and bias.

4.Making layers

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.inplanes, planes * block.expansion,
                          kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(planes * block.expansion),
            )

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes * block.expansion
        for i in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)

You can see that only the first convolutional layer in every stage has a residual layer.

ResNet

The layer is calculated as follows: the first convolutional layer with 7x7 kernel, the max pool layer and the next 48 blocks add up to 50. The network is like:

ResNet