Class Upsample2D

src/diffusers/models/upsampling.py:74–190

A 2D upsampling layer with an optional convolution. Parameters: channels (`int`): number of channels in the inputs and outputs. use_conv (`bool`, default `False`): option to use a convolution. use_conv_transpose (`bool`, default `False`):

class Upsample2D(nn.Module):
    """A 2D upsampling layer with an optional convolution.

    Parameters:
        channels (`int`):
            number of channels in the inputs and outputs.
        use_conv (`bool`, default `False`):
            option to use a convolution.
        use_conv_transpose (`bool`, default `False`):
            option to use a convolution transpose.
        out_channels (`int`, optional):
            number of output channels. Defaults to `channels`.
        name (`str`, default `conv`):
            name of the upsampling 2D layer.
    """

    def __init__(
        self,
        channels: int,
        use_conv: bool = False,
        use_conv_transpose: bool = False,
        out_channels: int | None = None,
        name: str = "conv",
        kernel_size: int | None = None,
        padding=1,
        norm_type=None,
        eps=None,
        elementwise_affine=None,
        bias=True,
        interpolate=True,
    ):
        super().__init__()
        self.channels = channels
        self.out_channels = out_channels or channels
        self.use_conv = use_conv
        self.use_conv_transpose = use_conv_transpose
        self.name = name
        self.interpolate = interpolate

        if norm_type == "ln_norm":
            self.norm = nn.LayerNorm(channels, eps, elementwise_affine)
        elif norm_type == "rms_norm":
            self.norm = RMSNorm(channels, eps, elementwise_affine)
        elif norm_type is None:
            self.norm = None
        else:
            raise ValueError(f"unknown norm_type: {norm_type}")

        conv = None
        if use_conv_transpose:
            if kernel_size is None:
                kernel_size = 4
            conv = nn.ConvTranspose2d(
                channels, self.out_channels, kernel_size=kernel_size, stride=2, padding=padding, bias=bias
            )
        elif use_conv:
            if kernel_size is None:
                kernel_size = 3