工程化

nn/nets/resnet.py

@@ -0,0 +1,83 @@
+import torch.nn as nn
+import torch.nn.functional as F
+from nn.const import num_classes
+
+
+class ResBlock(nn.Module):
+    def __init__(self, in_channel, out_channel, stride=1):
+        super(ResBlock, self).__init__()
+
+        self.layer = nn.Sequential(
+            nn.Conv2d(in_channel, out_channel, kernel_size=3, stride=stride, padding=1),
+            nn.BatchNorm2d(out_channel),
+            nn.ReLU(),
+            nn.Conv2d(out_channel, out_channel, kernel_size=3, stride=1, padding=1),
+            nn.BatchNorm2d(out_channel)
+        )
+
+        if in_channel != out_channel or stride > 1:
+            self.shortcut = nn.Sequential(
+                nn.Conv2d(in_channel, out_channel, kernel_size=1, stride=stride),
+                nn.BatchNorm2d(out_channel),
+            )
+        else:
+            self.shortcut = nn.Sequential()
+
+    def forward(self, x):
+        out = self.layer(x)
+        shortcut = self.shortcut(x)
+        out = out + shortcut
+        out = F.relu(out)
+        return out
+
+
+class ResNet(nn.Module):
+    def make_layer(self, block, out_channel, stride, num_block):
+        layer_list = []
+        for i in range(num_block):
+            if i == 0:
+                in_stride = stride
+            else:
+                in_stride = 1
+
+            layer_list.append(block(self.in_channel, out_channel, in_stride))
+            self.in_channel = out_channel
+            
+        return nn.Sequential(*layer_list)
+
+    def __init__(self):
+        super(ResNet, self).__init__()
+        self.in_channel = 32
+
+        self.conv1 = nn.Sequential(
+            nn.Conv2d(3, 32, kernel_size=3, stride=1, padding=1),
+            nn.BatchNorm2d(32),
+            nn.ReLU()
+        )
+
+        self.layer1 = self.make_layer(ResBlock, 64, 2, 2)   # 32 -> 64
+        self.layer2 = self.make_layer(ResBlock, 128, 2, 2)  # 64 -> 128
+        self.layer3 = self.make_layer(ResBlock, 256, 2, 2)  # 128 -> 256
+        self.layer4 = self.make_layer(ResBlock, 512, 2, 2)  # 256 -> 512
+
+        # 添加全局平均池化
+        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
+        self.fc = nn.Linear(512, num_classes)
+
+    def forward(self, x):
+        out = self.conv1(x)
+        out = self.layer1(out)
+        out = self.layer2(out)
+        out = self.layer3(out)
+        out = self.layer4(out)
+
+        # 使用全局平均池化
+        out = self.avgpool(out)
+        out = out.view(out.size(0), -1)
+        out = self.fc(out)
+
+        return out
+
+
+def resnet():
+    return ResNet()