| | |
| | |
| | |
| |
|
| | import torch.nn as nn |
| | import torch |
| |
|
| | class conv_block_nested(nn.Module): |
| | def __init__(self, in_ch, mid_ch, out_ch): |
| | super(conv_block_nested, self).__init__() |
| | self.activation = nn.ReLU(inplace=True) |
| | self.conv1 = nn.Conv2d(in_ch, mid_ch, kernel_size=3, padding=1, bias=True) |
| | self.bn1 = nn.BatchNorm2d(mid_ch) |
| | self.conv2 = nn.Conv2d(mid_ch, out_ch, kernel_size=3, padding=1, bias=True) |
| | self.bn2 = nn.BatchNorm2d(out_ch) |
| |
|
| | def forward(self, x): |
| | x = self.conv1(x) |
| | identity = x |
| | x = self.bn1(x) |
| | x = self.activation(x) |
| |
|
| | x = self.conv2(x) |
| | x = self.bn2(x) |
| | output = self.activation(x + identity) |
| | return output |
| |
|
| |
|
| | class up(nn.Module): |
| | def __init__(self, in_ch, bilinear=False): |
| | super(up, self).__init__() |
| |
|
| | if bilinear: |
| | self.up = nn.Upsample(scale_factor=2, |
| | mode='bilinear', |
| | align_corners=True) |
| | else: |
| | self.up = nn.ConvTranspose2d(in_ch, in_ch, 2, stride=2) |
| |
|
| | def forward(self, x): |
| | x = self.up(x) |
| | return x |
| |
|
| | class UNetpp(nn.Module): |
| | |
| | def __init__(self, in_ch=3): |
| | super(UNetpp, self).__init__() |
| | torch.nn.Module.dump_patches = True |
| | n1 = 32 |
| | filters = [n1, n1 * 2, n1 * 4, n1 * 8, n1 * 16] |
| |
|
| | self.pool = nn.MaxPool2d(kernel_size=2, stride=2) |
| |
|
| | self.conv0_0 = conv_block_nested(in_ch, filters[0], filters[0]) |
| | self.conv1_0 = conv_block_nested(filters[0], filters[1], filters[1]) |
| | self.Up1_0 = up(filters[1]) |
| | self.conv2_0 = conv_block_nested(filters[1], filters[2], filters[2]) |
| | self.Up2_0 = up(filters[2]) |
| | self.conv3_0 = conv_block_nested(filters[2], filters[3], filters[3]) |
| | self.Up3_0 = up(filters[3]) |
| | self.conv4_0 = conv_block_nested(filters[3], filters[4], filters[4]) |
| | self.Up4_0 = up(filters[4]) |
| |
|
| | self.conv0_1 = conv_block_nested(filters[0] * 2 + filters[1], filters[0], filters[0]) |
| | self.conv1_1 = conv_block_nested(filters[1] * 2 + filters[2], filters[1], filters[1]) |
| | self.Up1_1 = up(filters[1]) |
| | self.conv2_1 = conv_block_nested(filters[2] * 2 + filters[3], filters[2], filters[2]) |
| | self.Up2_1 = up(filters[2]) |
| | self.conv3_1 = conv_block_nested(filters[3] * 2 + filters[4], filters[3], filters[3]) |
| | self.Up3_1 = up(filters[3]) |
| |
|
| | self.conv0_2 = conv_block_nested(filters[0] * 3 + filters[1], filters[0], filters[0]) |
| | self.conv1_2 = conv_block_nested(filters[1] * 3 + filters[2], filters[1], filters[1]) |
| | self.Up1_2 = up(filters[1]) |
| | self.conv2_2 = conv_block_nested(filters[2] * 3 + filters[3], filters[2], filters[2]) |
| | self.Up2_2 = up(filters[2]) |
| |
|
| | self.conv0_3 = conv_block_nested(filters[0] * 4 + filters[1], filters[0], filters[0]) |
| | self.conv1_3 = conv_block_nested(filters[1] * 4 + filters[2], filters[1], filters[1]) |
| | self.Up1_3 = up(filters[1]) |
| |
|
| | self.conv0_4 = conv_block_nested(filters[0] * 5 + filters[1], filters[0], filters[0]) |
| |
|
| | for m in self.modules(): |
| | if isinstance(m, nn.Conv2d): |
| | nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu') |
| | elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)): |
| | nn.init.constant_(m.weight, 1) |
| | nn.init.constant_(m.bias, 0) |
| |
|
| |
|
| | def forward(self, xA, xB): |
| | '''xA''' |
| | x0_0A = self.conv0_0(xA) |
| | x1_0A = self.conv1_0(self.pool(x0_0A)) |
| | x2_0A = self.conv2_0(self.pool(x1_0A)) |
| | x3_0A = self.conv3_0(self.pool(x2_0A)) |
| | |
| | '''xB''' |
| | x0_0B = self.conv0_0(xB) |
| | x1_0B = self.conv1_0(self.pool(x0_0B)) |
| | x2_0B = self.conv2_0(self.pool(x1_0B)) |
| | x3_0B = self.conv3_0(self.pool(x2_0B)) |
| | x4_0B = self.conv4_0(self.pool(x3_0B)) |
| |
|
| | x0_1 = self.conv0_1(torch.cat([x0_0A, x0_0B, self.Up1_0(x1_0B)], 1)) |
| | x1_1 = self.conv1_1(torch.cat([x1_0A, x1_0B, self.Up2_0(x2_0B)], 1)) |
| | x0_2 = self.conv0_2(torch.cat([x0_0A, x0_0B, x0_1, self.Up1_1(x1_1)], 1)) |
| |
|
| |
|
| | x2_1 = self.conv2_1(torch.cat([x2_0A, x2_0B, self.Up3_0(x3_0B)], 1)) |
| | x1_2 = self.conv1_2(torch.cat([x1_0A, x1_0B, x1_1, self.Up2_1(x2_1)], 1)) |
| | x0_3 = self.conv0_3(torch.cat([x0_0A, x0_0B, x0_1, x0_2, self.Up1_2(x1_2)], 1)) |
| |
|
| | x3_1 = self.conv3_1(torch.cat([x3_0A, x3_0B, self.Up4_0(x4_0B)], 1)) |
| | x2_2 = self.conv2_2(torch.cat([x2_0A, x2_0B, x2_1, self.Up3_1(x3_1)], 1)) |
| | x1_3 = self.conv1_3(torch.cat([x1_0A, x1_0B, x1_1, x1_2, self.Up2_2(x2_2)], 1)) |
| | x0_4 = self.conv0_4(torch.cat([x0_0A, x0_0B, x0_1, x0_2, x0_3, self.Up1_3(x1_3)], 1)) |
| |
|
| | return [x0_1, x0_2, x0_3, x0_4] |
| |
|
| |
|
