17.1 DenseNet网络结构设计
import torch
from torch import nn
from torchsummary import summary
def conv_block(input_channels,num_channels):
net=nn.Sequential(nn.BatchNorm2d(input_channels),nn.ReLU(),nn.Conv2d(input_channels,num_channels,kernel_size=3,padding=1))
return net
def transition_block(inputs_channels,num_channels):
net=nn.Sequential(nn.BatchNorm2d(inputs_channels),nn.ReLU(),
nn.Conv2d(inputs_channels,num_channels,kernel_size=1),
nn.AvgPool2d(kernel_size=2,stride=2))
return net
class DenseBlock(nn.Module):
def __init__(self, num_convs,input_channels,num_channels):
super(DenseBlock,self).__init__()
layer=[]
for i in range(num_convs):
layer.append(conv_block(num_channels*i+input_channels,num_channels))
self.net=nn.Sequential(*layer)
def forward(self,X):
for blk in self.net:
Y=blk(X)
X=torch.cat((X,Y),dim=1)
return X
b1=nn.Sequential(nn.Conv2d(1,64,kernel_size=7,stride=2,padding=3),
nn.BatchNorm2d(64),nn.ReLU(),
nn.MaxPool2d(kernel_size=3,stride=2,padding=1))
num_channels,growth_rate=64,32
num_convs_in_dense_block=[4,4,4,4]
blks=[]
for i,num_convs in enumerate(num_convs_in_dense_block):
blks.append(DenseBlock(num_convs,num_channels,growth_rate))
num_channels+=num_convs*growth_rate
if i!=len(num_convs_in_dense_block)-1:
blks.append(transition_block(num_channels,num_channels//2))
num_channels=num_channels//2
model=nn.Sequential(b1,*blks,
nn.BatchNorm2d(num_channels),nn.ReLU(),
nn.AdaptiveAvgPool2d((1,1)),
nn.Flatten(),
nn.Linear(num_channels,10))
device=torch.device("cuda" if torch.cuda.is_available() else 'cpu')
model.to(device)
summary(model,input_size=(1,224,224),batch_size=64)
17.2 DenseNet网络实现Fashion-Mnist分类
import torch
import torchvision
from torch import nn
import matplotlib.pyplot as plt
from torchvision.transforms import transforms
from torch.utils.data import DataLoader
from tqdm import tqdm
from sklearn.metrics import accuracy_score
from torch.nn import functional as F
plt.rcParams['font.family']=['Times New Roman']
class Reshape(torch.nn.Module):
def forward(self,x):
return x.view(-1,1,28,28)
def plot_metrics(train_loss_list, train_acc_list, test_acc_list, title='Training Curve'):
epochs = range(1, len(train_loss_list) + 1)
plt.figure(figsize=(4, 3))
plt.plot(epochs, train_loss_list, label='Train Loss')
plt.plot(epochs, train_acc_list, label='Train Acc',linestyle='--')
plt.plot(epochs, test_acc_list, label='Test Acc', linestyle='--')
plt.xlabel('Epoch')
plt.ylabel('Value')
plt.title(title)
plt.legend()
plt.grid(True)
plt.tight_layout()
plt.show()
def train_model(model,train_data,test_data,num_epochs):
train_loss_list = []
train_acc_list = []
test_acc_list = []
for epoch in range(num_epochs):
total_loss=0
total_acc_sample=0
total_samples=0
loop=tqdm(train_data,desc=f"EPOCHS[{epoch+1}/{num_epochs}]")
for X,y in loop:
X=X.to(device)
y=y.to(device)
y_hat=model(X)
loss=CEloss(y_hat,y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
total_loss+=loss.item()*X.shape[0]
y_pred=y_hat.argmax(dim=1).detach().cpu().numpy()
y_true=y.detach().cpu().numpy()
total_acc_sample+=accuracy_score(y_pred,y_true)*X.shape[0]
total_samples+=X.shape[0]
test_acc_samples=0
test_samples=0
for X,y in test_data:
X=X.to(device)
y=y.to(device)
y_hat=model(X)
y_pred=y_hat.argmax(dim=1).detach().cpu().numpy()
y_true=y.detach().cpu().numpy()
test_acc_samples+=accuracy_score(y_pred,y_true)*X.shape[0]
test_samples+=X.shape[0]
avg_train_loss=total_loss/total_samples
avg_train_acc=total_acc_sample/total_samples
avg_test_acc=test_acc_samples/test_samples
train_loss_list.append(avg_train_loss)
train_acc_list.append(avg_train_acc)
test_acc_list.append(avg_test_acc)
print(f"Epoch {epoch+1}: Loss: {avg_train_loss:.4f},Trian Accuracy: {avg_train_acc:.4f},test Accuracy: {avg_test_acc:.4f}")
plot_metrics(train_loss_list, train_acc_list, test_acc_list)
return model
def init_weights(m):
if type(m) == nn.Linear or type(m) == nn.Conv2d:
nn.init.xavier_uniform_(m.weight)
def conv_block(input_channels,num_channels):
net=nn.Sequential(nn.BatchNorm2d(input_channels),nn.ReLU(),nn.Conv2d(input_channels,num_channels,kernel_size=3,padding=1))
return net
def transition_block(inputs_channels,num_channels):
net=nn.Sequential(nn.BatchNorm2d(inputs_channels),nn.ReLU(),
nn.Conv2d(inputs_channels,num_channels,kernel_size=1),
nn.AvgPool2d(kernel_size=2,stride=2))
return net
class DenseBlock(nn.Module):
def __init__(self, num_convs,input_channels,num_channels):
super(DenseBlock,self).__init__()
layer=[]
for i in range(num_convs):
layer.append(conv_block(num_channels*i+input_channels,num_channels))
self.net=nn.Sequential(*layer)
def forward(self,X):
for blk in self.net:
Y=blk(X)
X=torch.cat((X,Y),dim=1)
return X
b1=nn.Sequential(nn.Conv2d(1,64,kernel_size=7,stride=2,padding=3),
nn.BatchNorm2d(64),nn.ReLU(),
nn.MaxPool2d(kernel_size=3,stride=2,padding=1))
num_channels,growth_rate=64,32
num_convs_in_dense_block=[4,4,4,4]
blks=[]
for i,num_convs in enumerate(num_convs_in_dense_block):
blks.append(DenseBlock(num_convs,num_channels,growth_rate))
num_channels+=num_convs*growth_rate
if i!=len(num_convs_in_dense_block)-1:
blks.append(transition_block(num_channels,num_channels//2))
num_channels=num_channels//2
transforms=transforms.Compose([transforms.Resize(96),transforms.ToTensor(),transforms.Normalize((0.5,),(0.5,))])
train_img=torchvision.datasets.FashionMNIST(root="./data",train=True,transform=transforms,download=True)
test_img=torchvision.datasets.FashionMNIST(root="./data",train=False,transform=transforms,download=True)
train_data=DataLoader(train_img,batch_size=128,num_workers=4,shuffle=True)
test_data=DataLoader(test_img,batch_size=128,num_workers=4,shuffle=False)
device=torch.device("cuda" if torch.cuda.is_available() else 'cpu')
model=nn.Sequential(b1,*blks,
nn.BatchNorm2d(num_channels),nn.ReLU(),
nn.AdaptiveAvgPool2d((1,1)),
nn.Flatten(),
nn.Linear(num_channels,10))
model.to(device)
model.apply(init_weights)
optimizer=torch.optim.SGD(model.parameters(),lr=0.01,momentum=0.9)
CEloss=nn.CrossEntropyLoss()
model=train_model(model,train_data,test_data,num_epochs=10)
