| 100 | |
| 101 | |
| 102 | class Up(nn.Module): |
| 103 | def __init__(self, in_channels, out_channels, emb_dim=256): |
| 104 | super().__init__() |
| 105 | |
| 106 | self.up = nn.Upsample(scale_factor=2, mode="bilinear", align_corners=True) |
| 107 | self.conv = nn.Sequential( |
| 108 | DoubleConv(in_channels, in_channels, residual=True), |
| 109 | DoubleConv(in_channels, out_channels, in_channels // 2), |
| 110 | ) |
| 111 | |
| 112 | self.emb_layer = nn.Sequential( |
| 113 | nn.SiLU(), |
| 114 | nn.Linear( |
| 115 | emb_dim, |
| 116 | out_channels |
| 117 | ), |
| 118 | ) |
| 119 | |
| 120 | def forward(self, x, skip_x, t): |
| 121 | x = self.up(x) |
| 122 | x = torch.cat([skip_x, x], dim=1) |
| 123 | x = self.conv(x) |
| 124 | emb = self.emb_layer(t)[:, :, None, None].repeat(1, 1, x.shape[-2], x.shape[-1]) |
| 125 | return x + emb |
| 126 | |
| 127 | |
| 128 | class UNet(nn.Module): |