import torch
import torch.nn as nn
from torchvision.transforms.functional import resize
from ....utils import DECODER_REGISTRY
[docs]class DoubleConv(nn.Module):
"""
Module consisting of two convolution layers and activations
"""
def __init__(
self,
in_channels,
out_channels,
):
super(DoubleConv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(in_channels, out_channels, 3, 1, 1, bias=False),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True),
nn.Conv2d(out_channels, out_channels, 3, 1, 1, bias=False),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True),
)
[docs] def forward(self, x):
return self.conv(x)
[docs]@DECODER_REGISTRY.register()
class SegmentationHead(nn.Module):
"""
U-net like up-sampling block
"""
def __init__(
self,
out_channels=1,
embed_dims=[64, 128, 256, 512],
):
super(SegmentationHead, self).__init__()
self.ups = nn.ModuleList()
self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
for feature in reversed(embed_dims):
self.ups.append(
nn.ConvTranspose2d(
feature * 2,
feature,
kernel_size=2,
stride=2,
)
)
self.ups.append(DoubleConv(feature * 2, feature))
self.bottleneck = DoubleConv(embed_dims[-1], embed_dims[-1] * 2)
self.conv1 = nn.Conv2d(embed_dims[0], out_channels, kernel_size=1)
self.conv2 = nn.ConvTranspose2d(
out_channels, out_channels, kernel_size=4, stride=4
)
[docs] def forward(self, skip_connections):
x = self.bottleneck(skip_connections[-1])
skip_connections = skip_connections[::-1]
for idx in range(0, len(self.ups), 2):
x = self.ups[idx](x)
skip_connection = skip_connections[idx // 2]
if x.shape != skip_connection.shape:
x = resize(x, size=skip_connection.shape[2:])
concat_skip = torch.cat((skip_connection, x), dim=1)
x = self.ups[idx + 1](concat_skip)
x = self.conv1(x)
return self.conv2(x)