Reposted from ChatGPT, non-original, used to compare your own models. Not much nonsense, just upload the code~
- U-Net model improvements:
- Use a deeper network structure, such as adding more convolutional layers and pooling layers, to improve the expressiveness and generalization ability of the model.
- Use different loss functions, such as Focal Loss, Dice Loss, etc., to improve the accuracy of the model for edge information.
- Data enhancement processing, such as rotation, flipping, scaling, etc., to increase the diversity of data and improve the robustness of the model.
The following is the code of the U-Net model:
import torch.nn as nn
class DoubleConv(nn.Module):
def __init__(self, in_channels, out_channels):
super(DoubleConv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(in_channels, out_channels, 3, padding=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True),
nn.Conv2d(out_channels, out_channels, 3, padding=1),
nn.BatchNorm2d(out_channels),
nn.ReLU(inplace=True)
)
def forward(self, x):
return self.conv(x)
class UNet(nn.Module):
def __init__(self, in_channels=1, out_channels=1, features=[64, 128, 256, 512]):
super(UNet, self).__init__()
self.ups = nn.ModuleList()
self.downs = nn.ModuleList()
self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
# Down part of U-Net
for feature in features:
self.downs.append(DoubleConv(in_channels, feature))
in_channels = feature
# Up part of U-Net
for feature in reversed(features):
self.ups.append(nn.ConvTranspose2d(feature*2, feature, kernel_size=2, stride=2))
self.ups.append(DoubleConv(feature*2, feature))
self. bottleneck = DoubleConv(features[-1], features[-1]*2)
self.final_conv = nn.Conv2d(features[0], out_channels, kernel_size=1)
def forward(self, x):
skip_connections = []
for down in self. downs:
x = down(x)
skip_connections.append(x)
x = self. pool(x)
x = self. bottleneck(x)
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 = TF.resize(x, size=skip_connection.shape[2:])
concat_skip = torch.cat((skip_connection, x), dim=1)
x = self.ups[idx + 1](concat_skip)
return self. final_conv(x)
2. DeepLabv3 + model improvements:
- Add the spatial pyramid pooling module to improve the model’s ability to capture information of different scales.
- The deformable convolutional layer is used to increase the receptive field of the model and improve the adaptability of the model to the target shape.
- Multi-scale training and testing are used to improve the model’s ability to detect objects of different scales.
Below is the code for the DeepLabv3+ model:
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils import model_zoo
model_urls = {
'resnet50': 'https://download.pytorch.org/models/resnet50-19c8e357.pth',
'resnet101': 'https://download.pytorch.org/models/resnet101-5d3b4d8f.pth',
}
class ASPP(nn.Module):
def __init__(self, in_channels, out_channels=256, rates=[6, 12, 18]):
super(ASPP, self).__init__()
self.conv1x1 = nn.Conv2d(in_channels, out_channels, kernel_size=1)
self.atrous_conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=rates[0], dilation=rates[0])
self.atrous_conv2 = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=rates[1], dilation=rates[1])
self.atrous_conv3 = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=rates[2], dilation=rates[2])
self.pool = nn.Sequential(
nn.AdaptiveAvgPool2d(1),
nn.Conv2d(in_channels, out_channels, kernel_size=1)
)
self.conv = nn.Conv2d(out_channels*5, out_channels, kernel_size=1)
def forward(self, x):
feature_map = self.conv1x1(x)
atrous_1 = self.atrous_conv1(x)
atrous_2 = self.atrous_conv2(x)
atrous_3 = self.atrous_conv3(x)
pool = F.interpolate(self.pool(x), size=feature_map.shape[2:], mode='bilinear', align_corners=True)
x = torch.cat((feature_map, atrous_1, atrous_2, atrous_3, pool), dim=1)
return self.conv(x)
class DeepLabv3Plus(nn.Module):
def __init__(self, in_channels=3, out_channels=21, backbone='resnet50', pretrained=True):
super(DeepLabv3Plus, self).__init__()
if backbone == 'resnet50':
resnet = models.resnet50(pretrained=pretrained)
channels = 2048
elif backbone == 'resnet101':
resnet = models.resnet101(pretrained=pretrained)
channels = 2048
self.conv1 = resnet.conv1
self.bn1 = resnet.bn1
self.relu = resnet.relu
self.maxpool = resnet.maxpool
self.layer1 = resnet.layer1
self.layer2 = resnet.layer2
self.layer3 = resnet.layer3
self.layer4 = resnet.layer4
self.aspp = ASPP(channels)
self.up_conv1 = nn.ConvTranspose2d(channels//2, channels//4, kernel_size=4, stride=2, padding=1)
self.up_conv2 = nn.ConvTranspose2d(channels//4, channels//8, kernel_size=4, stride=2, padding=1)
self.up_conv3 = nn.ConvTranspose2d(channels//8, channels//16, kernel_size=4, stride=2, padding=1)
self.final_conv = nn.Conv2d(channels//16, out_channels, kernel_size=1)
def forward(self, x):
x_size = x.size()
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self. maxpool(x)
x = self. layer1(x)
x = self. layer2(x)
x = self. layer3(x)
x = self. layer4(x)
x = self.aspp(x)
x = self.up_conv1(x)
x = F.interpolate(x, size=self.layer3(x).size()[2:], mode='bilinear', align_corners=True)
x = torch.cat((x, self.layer3(x)), dim=1)
x = self.up_conv2(x)
x = F.interpolate(x, size=self.layer2(x).size()[2:], mode='bilinear', align_corners=True)
x = torch.cat((x, self.layer2(x)), dim=1)
x = self.up_conv3(x)
x = F.interpolate(x, size=self.layer1(x).size()[2:], mode='bilinear', align_corners=True)
x = torch.cat((x, self.layer1(x)), dim=1)
x = self. final_conv(x)
x = F. interpolate(x, size=x_size[2:], mode='bilinear', align_corners=True)
return x