Skip to content

film_residual_block

film_residual_block

FiLMedResBlock 

FiLMedResBlock(in_channels, out_channels, conditioning_dimension, stride=1, downsample=None)

Bases: Module

ResNet residual block with FiLM conditioning.

Source code in src/versatil/models/layers/modulation/film_residual_block.py 
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
def __init__(
    self,
    in_channels: int,
    out_channels: int,
    conditioning_dimension: int,
    stride: int = 1,
    downsample: nn.Module | None = None,
):
    super().__init__()

    self.conv1 = nn.Conv2d(
        in_channels,
        out_channels,
        kernel_size=3,
        stride=stride,
        padding=1,
        bias=False,
    )
    self.bn1 = nn.BatchNorm2d(out_channels)
    self.film1 = ConditionalModulation(
        conditioning_dimension=conditioning_dimension,
        feature_dim=out_channels,
        use_shift=True,
        activation=ActivationFunction.LINEAR.value,
        init_strategy="zero",
    )

    self.conv2 = nn.Conv2d(
        out_channels, out_channels, kernel_size=3, stride=1, padding=1, bias=False
    )
    self.bn2 = nn.BatchNorm2d(out_channels)
    self.film2 = ConditionalModulation(
        conditioning_dimension=conditioning_dimension,
        feature_dim=out_channels,
        use_shift=True,
        activation=ActivationFunction.LINEAR.value,
        init_strategy="zero",
    )

    self.relu = nn.ReLU()
    self.downsample = downsample

forward 

forward(x, condition)

Apply the FiLM-modulated residual convolution block.

Source code in src/versatil/models/layers/modulation/film_residual_block.py 
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
def forward(self, x: torch.Tensor, condition: torch.Tensor) -> torch.Tensor:
    """Apply the FiLM-modulated residual convolution block."""
    identity = x if self.downsample is None else self.downsample(x)

    out = self.conv1(x)
    out = self.bn1(out)
    out, _ = self.film1(out, condition)
    out = self.relu(out)

    out = self.conv2(out)
    out = self.bn2(out)
    out, _ = self.film2(out, condition)

    out = out + identity
    result: torch.Tensor = self.relu(out)

    return result