这个内容是将随便做了一个网络结构，然后简单的训练几次，生成模型，并且存储起来，主要是为了学习获得pytorch中的BatchNorm2d层的各个特征图的平均值和方差。代码如下：
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import datasets,transforms
from torch.optim import lr_scheduler
import torch.optim as optim
class VGG(nn.Module):
def __init__(self):
super(VGG,self).__init__()
self.conv1 = nn.Conv2d(3,64,3,padding=(1,1))
self.bn1 = nn.BatchNorm2d(64)
self.maxpool1 = nn.MaxPool2d((2,2))
self.conv2 = nn.Conv2d(64,128,3,padding=(1,1))
# self.bn2 = nn.BatchNorm2d(128)
self.maxpool2 = nn.MaxPool2d((2,2))
self.conv3 = nn.Conv2d(128,256,3,padding=(1,1))
# self.bn3 = nn.BatchNorm2d(256)
self.maxpool3 = nn.MaxPool2d((2,2))
self.fc1 = nn.Linear(256*16*8,4096)
self.fc2 = nn.Linear(4096,1000)
self.fc3 = nn.Linear(1000,10)
def forward(self,x):
in_size = x.size(0)
out = self.conv1(x)
out = self.bn1(out)
out = F.relu(out)
out = self.maxpool1(out)
out = self.conv2(out)
out = F.relu(out)
out = self.maxpool2(out)
out = self.conv3(out)
out = F.relu(out)
out = self.maxpool3(out)
out = out.view(out.size(0),-1)
out = self.fc1(out)
out = F.relu(out)
out = self.fc2(out)
out = F.relu(out)
out = self.fc3(out)
return out
transform_train_list = transforms.Compose([
transforms.Resize( (256,128),interpolation=3 ),
transforms.RandomCrop((128,64)),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
train_dataset = datasets.ImageFolder('./train',transform_train_list)
dataloaders = torch.utils.data.DataLoader(train_dataset,batch_size=2,num_workers=0)
dataset_size = len(train_dataset)
class_names = train_dataset.classes
print(dataset_size)
print(class_names)
net=VGG()
optimizer = optim.SGD(net.parameters(),lr=0.001,momentum=0.9)
criterion = nn.CrossEntropyLoss()
for epoch in range(2):#训练模型
print(epoch)
net.train(True)
running_loss = 0.0
running_corrects = 0.0
for data in dataloaders:
inputs,labels = data
now_batch,c,h,w = inputs.shape
optimizer.zero_grad()
outputs = net(inputs)
# print(outputs)
_,preds = torch.max(outputs.data,1)
loss = criterion(outputs,labels)
loss.backward()
optimizer.step()
running_loss = running_loss + loss.item() * now_batch
running_corrects += float( torch.sum( preds == labels.data ) )
epoch_loss = running_loss/dataset_size
epoch_acc = running_corrects/dataset_size
print(epoch_loss)
print(epoch_acc)
torch.save(net.cpu().state_dict(),'first.pth') ##将训练好的模型保存起来
net = VGG()
net.load_state_dict( torch.load('first.pth') )
net.eval() #产生一个模型并且加载已经训练好的模型的参数
# for data in dataloaders:
# inputs,labels = data
# # print(inputs)
# print(labels)
# outputs = net(inputs)
# print(outputs)
# break
m = VGG()
# m.eval()
m.load_state_dict( torch.load('second.pth') )
print(m.bn1.running_mean.size()) ##获得一共有多少个mean 要是想获得var只要将mean改为var即可
print(m.bn1.running_mean.data[0])
print(m.bn1.running_mean.data[1])
print(m.bn1.running_mean.data[2])
print(m.bn1.running_var.data[0])
print(type(m.bn1.running_mean.data[0]))
m.bn1.running_mean.data[0] = m.bn1.running_mean.data[2] ##可以对模型参数进行更改，然后保存更改后的模型
m.bn1.running_mean.data[1] = m.bn1.running_mean.data[2]
torch.save(m.cpu().state_dict(),'second.pth')
对于输入到BatchNorm2d层的数据格式为(batch_size,channels_size,h,w)，channels_size为多少，就会生成多少个mean和var。
举个例子，如果输入的数据是batch_size=16,channels_size=64,h=32,w=16,则每对mean和var都是 16张某一个特征图中的所有数据的mean和var
