model.apply(weights_init_normal)

model.apply(weights_init_normal)方法

应用把方法应用于每一个module,这里意思是进行初始化

1. def weights_init_normal(m): 
   

   ---

2. classname = m.__class__.__name__ 
   

   ---

3. if classname.find("Conv") != -1: 
   

   ---

4. torch.nn.init.normal_(m.weight.data, 0.0, 0.02) 
   

   ---

5. elif classname.find("BatchNorm2d") != -1: 
   

   ---

6. torch.nn.init.normal_(m.weight.data, 1.0, 0.02) 
   

   ---

7. torch.nn.init.constant_(m.bias.data, 0.0) 
   

   ---

这里的意思是选择module是conv或者是batchNorm2d的层进行初始化