Transposed convolution in_channel and out_channels for DCGAN generator network

Question

I'm looking at this code for the generator network of a DCGAN based on the original DCGAN paper.

z_size = 100

num_g_filters = 64

netG = nn.Sequential(
    nn.ConvTranspose2d(
        in_channels=z_size,
        out_channels=num_g_filters * 8,
        kernel_size=4,
        stride=1,
        padding=0,
        bias=False
    ),
    nn.BatchNorm2d(num_features=num_g_filters * 8),
    nn.ReLU(inplace=True),

    nn.ConvTranspose2d(
        in_channels=num_g_filters * 8,
        out_channels=num_g_filters * 4,
        kernel_size=4,
        stride=2,
        padding=1,
        bias=False
    ),
    nn.BatchNorm2d(num_features=num_g_filters * 4),
    nn.ReLU(inplace=True),

    nn.ConvTranspose2d(
        in_channels=num_g_filters * 4,
        out_channels=num_g_filters * 2,
        kernel_size=4,
        stride=2,
        padding=1,
        bias=False
    ),
    nn.BatchNorm2d(num_features=num_g_filters * 2),
    nn.ReLU(inplace=True),

    nn.ConvTranspose2d(
        in_channels=num_g_filters * 2,
        out_channels=num_g_filters,
        kernel_size=4,
        stride=2,
        padding=1,
        bias=False
    ),
    nn.BatchNorm2d(num_features=num_g_filters),
    nn.ReLU(inplace=True),

    nn.ConvTranspose2d(
        in_channels=num_g_filters,
        out_channels=num_channels,
        kernel_size=4,
        stride=2,
        padding=1, 
        bias=False
    ),
    nn.Tanh()
    )

Where does the z_size vector get projected and reshaped from a length of 100 to 4 x 4 x 1024? The first transposed convolution takes in_channels = 100 and outputs out_channels = 512 seemingly skipping the projecting and reshaping step. There also seems to be an additional transposed convolution that takes in_channels = 64 * 2 = 128 and outputs out_channels = 64 which I don't see in the visual of the network. In the diagram, the last transposed convolution takes in 128 channels and outputs 3 channels.