(
self,
*,
in_channels: int,
out_channels: int | None = None,
conv_shortcut: bool = False,
dropout: float = 0.0,
temb_channels: int = 512,
groups: int = 32,
groups_out: int | None = None,
eps: float = 1e-6,
non_linearity: str = "swish",
time_embedding_norm: str = "ada_group", # ada_group, spatial
output_scale_factor: float = 1.0,
use_in_shortcut: bool | None = None,
up: bool = False,
down: bool = False,
conv_shortcut_bias: bool = True,
conv_2d_out_channels: int | None = None,
)
| 71 | """ |
| 72 | |
| 73 | def __init__( |
| 74 | self, |
| 75 | *, |
| 76 | in_channels: int, |
| 77 | out_channels: int | None = None, |
| 78 | conv_shortcut: bool = False, |
| 79 | dropout: float = 0.0, |
| 80 | temb_channels: int = 512, |
| 81 | groups: int = 32, |
| 82 | groups_out: int | None = None, |
| 83 | eps: float = 1e-6, |
| 84 | non_linearity: str = "swish", |
| 85 | time_embedding_norm: str = "ada_group", # ada_group, spatial |
| 86 | output_scale_factor: float = 1.0, |
| 87 | use_in_shortcut: bool | None = None, |
| 88 | up: bool = False, |
| 89 | down: bool = False, |
| 90 | conv_shortcut_bias: bool = True, |
| 91 | conv_2d_out_channels: int | None = None, |
| 92 | ): |
| 93 | super().__init__() |
| 94 | self.in_channels = in_channels |
| 95 | out_channels = in_channels if out_channels is None else out_channels |
| 96 | self.out_channels = out_channels |
| 97 | self.use_conv_shortcut = conv_shortcut |
| 98 | self.up = up |
| 99 | self.down = down |
| 100 | self.output_scale_factor = output_scale_factor |
| 101 | self.time_embedding_norm = time_embedding_norm |
| 102 | |
| 103 | if groups_out is None: |
| 104 | groups_out = groups |
| 105 | |
| 106 | if self.time_embedding_norm == "ada_group": # ada_group |
| 107 | self.norm1 = AdaGroupNorm(temb_channels, in_channels, groups, eps=eps) |
| 108 | elif self.time_embedding_norm == "spatial": |
| 109 | self.norm1 = SpatialNorm(in_channels, temb_channels) |
| 110 | else: |
| 111 | raise ValueError(f" unsupported time_embedding_norm: {self.time_embedding_norm}") |
| 112 | |
| 113 | self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=1, padding=1) |
| 114 | |
| 115 | if self.time_embedding_norm == "ada_group": # ada_group |
| 116 | self.norm2 = AdaGroupNorm(temb_channels, out_channels, groups_out, eps=eps) |
| 117 | elif self.time_embedding_norm == "spatial": # spatial |
| 118 | self.norm2 = SpatialNorm(out_channels, temb_channels) |
| 119 | else: |
| 120 | raise ValueError(f" unsupported time_embedding_norm: {self.time_embedding_norm}") |
| 121 | |
| 122 | self.dropout = torch.nn.Dropout(dropout) |
| 123 | |
| 124 | conv_2d_out_channels = conv_2d_out_channels or out_channels |
| 125 | self.conv2 = nn.Conv2d(out_channels, conv_2d_out_channels, kernel_size=3, stride=1, padding=1) |
| 126 | |
| 127 | self.nonlinearity = get_activation(non_linearity) |
| 128 | |
| 129 | self.upsample = self.downsample = None |
| 130 | if self.up: |
nothing calls this directly
no test coverage detected