A 1D downsampling 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. out_channels (`int`, optional): number o
| 25 | |
| 26 | |
| 27 | class Downsample1D(nn.Module): |
| 28 | """A 1D downsampling layer with an optional convolution. |
| 29 | |
| 30 | Parameters: |
| 31 | channels (`int`): |
| 32 | number of channels in the inputs and outputs. |
| 33 | use_conv (`bool`, default `False`): |
| 34 | option to use a convolution. |
| 35 | out_channels (`int`, optional): |
| 36 | number of output channels. Defaults to `channels`. |
| 37 | padding (`int`, default `1`): |
| 38 | padding for the convolution. |
| 39 | name (`str`, default `conv`): |
| 40 | name of the downsampling 1D layer. |
| 41 | """ |
| 42 | |
| 43 | def __init__( |
| 44 | self, |
| 45 | channels: int, |
| 46 | use_conv: bool = False, |
| 47 | out_channels: Optional[int] = None, |
| 48 | padding: int = 1, |
| 49 | name: str = "conv", |
| 50 | ): |
| 51 | super().__init__() |
| 52 | self.channels = channels |
| 53 | self.out_channels = out_channels or channels |
| 54 | self.use_conv = use_conv |
| 55 | self.padding = padding |
| 56 | stride = 2 |
| 57 | self.name = name |
| 58 | |
| 59 | if use_conv: |
| 60 | self.conv = nn.Conv1d(self.channels, self.out_channels, 3, stride=stride, padding=padding) |
| 61 | else: |
| 62 | assert self.channels == self.out_channels |
| 63 | self.conv = nn.AvgPool1d(kernel_size=stride, stride=stride) |
| 64 | |
| 65 | def forward(self, inputs: torch.Tensor) -> torch.Tensor: |
| 66 | assert inputs.shape[1] == self.channels |
| 67 | return self.conv(inputs) |
| 68 | |
| 69 | |
| 70 | class Downsample2D(nn.Module): |