| import torch
|
| import torch.nn as nn
|
|
|
| class UNetBlock(nn.Module):
|
| def __init__(self, in_channels, out_channels, down=True, use_dropout=False):
|
| super().__init__()
|
| if down:
|
| self.block = nn.Sequential(
|
| nn.Conv2d(in_channels, out_channels, 4, 2, 1, bias=False),
|
| nn.BatchNorm2d(out_channels),
|
| nn.LeakyReLU(0.2, inplace=True)
|
| )
|
| else:
|
| self.block = nn.Sequential(
|
| nn.ConvTranspose2d(in_channels, out_channels, 4, 2, 1, bias=False),
|
| nn.BatchNorm2d(out_channels),
|
| nn.ReLU(inplace=True)
|
| )
|
| self.use_dropout = use_dropout
|
| if use_dropout:
|
| self.dropout = nn.Dropout(0.5)
|
|
|
| def forward(self, x):
|
| x = self.block(x)
|
| return self.dropout(x) if self.use_dropout else x
|
|
|
|
|
| class UNetGenerator(nn.Module):
|
| def __init__(self, in_channels=3, out_channels=3):
|
| super().__init__()
|
|
|
| self.down1 = UNetBlock(in_channels, 64, down=True, use_dropout=False)
|
| self.down2 = UNetBlock(64, 128)
|
| self.down3 = UNetBlock(128, 256)
|
| self.down4 = UNetBlock(256, 512)
|
| self.down5 = UNetBlock(512, 512)
|
| self.down6 = UNetBlock(512, 512)
|
| self.down7 = UNetBlock(512, 512)
|
| self.down8 = nn.Sequential(
|
| nn.Conv2d(512, 512, 4, 2, 1), nn.ReLU()
|
| )
|
|
|
|
|
| self.up1 = UNetBlock(512, 512, down=False, use_dropout=True)
|
| self.up2 = UNetBlock(1024, 512, down=False, use_dropout=True)
|
| self.up3 = UNetBlock(1024, 512, down=False, use_dropout=True)
|
| self.up4 = UNetBlock(1024, 512, down=False)
|
| self.up5 = UNetBlock(1024, 256, down=False)
|
| self.up6 = UNetBlock(512, 128, down=False)
|
| self.up7 = UNetBlock(256, 64, down=False)
|
|
|
| self.final = nn.Sequential(
|
| nn.ConvTranspose2d(128, out_channels, 4, 2, 1),
|
| nn.Tanh()
|
| )
|
|
|
| def forward(self, x):
|
|
|
| d1 = self.down1(x)
|
| d2 = self.down2(d1)
|
| d3 = self.down3(d2)
|
| d4 = self.down4(d3)
|
| d5 = self.down5(d4)
|
| d6 = self.down6(d5)
|
| d7 = self.down7(d6)
|
| bottleneck = self.down8(d7)
|
|
|
|
|
| u1 = self.up1(bottleneck)
|
| u2 = self.up2(torch.cat([u1, d7], dim=1))
|
| u3 = self.up3(torch.cat([u2, d6], dim=1))
|
| u4 = self.up4(torch.cat([u3, d5], dim=1))
|
| u5 = self.up5(torch.cat([u4, d4], dim=1))
|
| u6 = self.up6(torch.cat([u5, d3], dim=1))
|
| u7 = self.up7(torch.cat([u6, d2], dim=1))
|
| out = self.final(torch.cat([u7, d1], dim=1))
|
| return out
|
| |
