取消初始化权重函数

Author: Kedreamix

CIFAR10_code/nets/AlexNet.py

@@ -7,7 +7,7 @@
 
 # 定义2012的AlexNet
 class AlexNet(nn.Module): 
-    def __init__(self,num_classes=10, init_weights=True):
+    def __init__(self,num_classes=10):
         super(AlexNet,self).__init__()
         # 五个卷积层 输入 32 * 32 * 3
         self.conv1 = nn.Sequential(
@@ -43,8 +43,7 @@ def __init__(self,num_classes=10, init_weights=True):
             nn.ReLU(),
             nn.Linear(84,num_classes)
         )
-        if init_weights:
-            self._initialize_weights()
+
     def forward(self,x):
         x = self.conv1(x)
         x = self.conv2(x)
@@ -55,18 +54,6 @@ def forward(self,x):
         x = self.fc(x)
         return x
 
-    def _initialize_weights(self):
-        for m in self.modules():
-            if isinstance(m, nn.Conv2d):
-                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
-                if m.bias is not None:
-                    nn.init.constant_(m.bias, 0)
-            elif isinstance(m, nn.BatchNorm2d):
-                nn.init.constant_(m.weight, 1)
-                nn.init.constant_(m.bias, 0)
-            elif isinstance(m, nn.Linear):
-                nn.init.normal_(m.weight, 0, 0.01)
-                nn.init.constant_(m.bias, 0)
 def test():
     net = AlexNet()
     x = torch.randn(2,3,32,32)

CIFAR10_code/nets/LeNet5.py

@@ -5,7 +5,7 @@
 import torch.nn as nn
 
 class LeNet5(nn.Module):
-    def __init__(self, num_classes = 10, init_weights=True):
+    def __init__(self, num_classes = 10):
         super(LeNet5,self).__init__()
         self.conv1 = nn.Sequential(
             # 输入 32x32x3 -> 28x28x6 (32-5)/1 + 1=28
@@ -30,8 +30,7 @@ def __init__(self, num_classes = 10, init_weights=True):
             nn.ReLU(),
             nn.Linear(84,num_classes)
         )
-        if init_weights:
-            self._initialize_weights()
+        
     def forward(self,x):
         x = self.conv1(x)
         x = self.conv2(x)
@@ -40,18 +39,7 @@ def forward(self,x):
         x = self.fc(x)
         return x
 
-    def _initialize_weights(self):
-        for m in self.modules():
-            if isinstance(m, nn.Conv2d):
-                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
-                if m.bias is not None:
-                    nn.init.constant_(m.bias, 0)
-            elif isinstance(m, nn.BatchNorm2d):
-                nn.init.constant_(m.weight, 1)
-                nn.init.constant_(m.bias, 0)
-            elif isinstance(m, nn.Linear):
-                nn.init.normal_(m.weight, 0, 0.01)
-                nn.init.constant_(m.bias, 0)    
+
 
 def test():
     net = LeNet5()

CIFAR10_code/nets/MobileNetv2.py

@@ -68,18 +68,6 @@ def _make_layers(self, in_channels):
                 in_channels = out_channels
         return nn.Sequential(*layers)
 
-    def init_weight(self):
-        for w in self.modules():
-            if isinstance(w, nn.Conv2d):
-                nn.init.kaiming_normal_(w.weight, mode='fan_out')
-                if w.bias is not None:
-                    nn.init.zeros_(w.bias)
-            elif isinstance(w, nn.BatchNorm2d):
-                nn.init.ones_(w.weight)
-                nn.init.zeros_(w.bias)
-            elif isinstance(w, nn.Linear):
-                nn.init.normal_(w.weight, 0, 0.01)
-                nn.init.zeros_(w.bias)    
 
     def forward(self, x):
         out = self.relu6(self.bn1(self.conv1(x)))

CIFAR10_code/train.py

@@ -1,5 +1,4 @@
 '''Train CIFAR10 with PyTorch.'''
-import imp
 import torch
 import torch.nn as nn
 import torch.optim as optim