| import torch |
| import torch.nn as nn |
|
|
|
|
| class ResidualBlock(nn.Module): |
| def __init__(self, in_channels, out_channels, dropout_rate=0.2): |
| super(ResidualBlock, self).__init__() |
| self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1) |
| self.bn1 = nn.BatchNorm2d(out_channels) |
| self.dropout1 = nn.Dropout2d(p=dropout_rate) |
| self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1) |
| self.bn2 = nn.BatchNorm2d(out_channels) |
| self.dropout2 = nn.Dropout2d(p=dropout_rate) |
|
|
| self.skip_connection = nn.Sequential() |
| if in_channels != out_channels: |
| self.skip_connection = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=1, padding=0) |
|
|
| def forward(self, x): |
| residual = self.skip_connection(x) |
| out = nn.functional.relu(self.bn1(self.conv1(x))) |
| out = self.dropout1(out) |
| out = self.bn2(self.conv2(out)) |
| out = self.dropout2(out) |
| out += residual |
| out = nn.functional.relu(out) |
| return out |
|
|
|
|
| class MyModel(nn.Module): |
| def __init__(self, num_classes=100, dropout_rate=0.2): |
| super(MyModel, self).__init__() |
| self.dropout_rate = dropout_rate |
|
|
| self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3) |
| self.bn1 = nn.BatchNorm2d(64) |
| self.pool1 = nn.MaxPool2d(kernel_size=3, stride=2, padding=1) |
| self.dropout1 = nn.Dropout2d(p=self.dropout_rate) |
|
|
| |
| self.block1 = self._resnet_layers(64, 128, num_blocks=4) |
| self.block2 = self._resnet_layers(128, 256, num_blocks=4) |
| self.block3 = self._resnet_layers(256, 512, num_blocks=4) |
|
|
| self.global_avg_pool = nn.AdaptiveAvgPool2d(1) |
| self.dropout2 = nn.Dropout(p=self.dropout_rate) |
|
|
| |
| self.fc = nn.Linear(512, num_classes) |
|
|
| self.features = nn.Sequential( |
| self.conv1, |
| self.bn1, |
| nn.ReLU(), |
| self.pool1, |
| self.dropout1, |
| self.block1, |
| self.block2, |
| self.block3, |
| self.global_avg_pool, |
| self.dropout2 |
| ) |
|
|
| @staticmethod |
| def _resnet_layers(in_channels, out_channels, num_blocks): |
| return nn.Sequential( |
| ResidualBlock(in_channels, out_channels, dropout_rate=0.2), |
| *[ResidualBlock(out_channels, out_channels, dropout_rate=0.2) for _ in range(num_blocks)] |
| ) |
|
|
| def forward(self, x): |
| x = self.features(x) |
| x = torch.flatten(x, 1) |
| x = self.fc(x) |
| return x |
|
|
