Inception Module实现
- 一、Inception网络简介
- 二、Inception Module
- 三、使用该Inception Module实现MNIST
一、Inception网络简介
Inception 又叫GoogleNet,一般来说CNN模型提升网络性能的方法是增加深度(层数)或宽度(层的通道数),但这样进行网络设计一般会带来巨量的计算开销。GoogleNet借鉴了诸多前人的观点与经验(尤其是Network in Network中使用1x1 conv及AvgPool的idea),如果每个Inception module里的计算都由各自的1x1 conv来隔离,就不会像传统CNN深度模型那样随着深度增加其计算量也指数级增加,如下图所示,通过加入1x1 conv,计算量得到了明显的减少。
![](https://img.laitimes.com/img/__Qf2AjLwojIjJCLyojI0JCLiAzNfRHLGZkRGZkRfJ3bs92YsYTMfVmepNHLw0EVOd3Zq10MNpHW4Z0MMBjVtJWd0ckW65UbM5WOHJWa5kHT20ESjBjUIF2X0hXZ0xCMx81dvRWYoNHLrdEZwZ1Rh5WNXp1bwNjW1ZUba9VZwlHdssmch1mclRXY39CXldWYtlWPzNXZj9mcw1ycz9WL49zZuBnL5MTM4ITN0UTMxADNwEjMwIzLc52YucWbp5GZzNmLn9Gbi1yZtl2Lc9CX6MHc0RHaiojIsJye.png)
二、Inception Module
对上图所示Inception Module 进行实现
代码如下:
class InceptionA(torch.nn.Module):
def __init__(self, in_channels):
super(InceptionA, self).__init__()
self.branch1x1 = torch.nn.Conv2d(in_channels, 16, kernel_size=(1, 1))
self.branch5x5_1 = torch.nn.Conv2d(in_channels, 16, kernel_size=(1, 1))
# 因为kernel_size=(5, 5),为了保证图像宽高不改变,将padding设置为2,5/2=2
self.branch5x5_2 = torch.nn.Conv2d(16, 24, kernel_size=(5, 5), padding=(2, 2))
self.branch3x3_1 = torch.nn.Conv2d(in_channels, 16, kernel_size=(1, 1))
# 因为kernel_size=(3, 3),为了保证图像宽高不改变,将padding设置为1,3/2=1
self.branch3x3_2 = torch.nn.Conv2d(16, 24, kernel_size=(3, 3), padding=(1, 1))
# 因为kernel_size=(3, 3),为了保证图像宽高不改变,将padding设置为1,3/2=1
self.branch3x3_3 = torch.nn.Conv2d(24, 24, kernel_size=(3, 3), padding=(1, 1))
self.branch_pool = torch.nn.Conv2d(in_channels, 24, kernel_size=(1, 1))
def forward(self, x):
branch1x1 = self.branch1x1(x)
branch5x5 = self.branch5x5_1(x)
branch5x5 = self.branch5x5_2(branch5x5)
branch3x3 = self.branch3x3_1(x)
branch3x3 = self.branch3x3_2(branch3x3)
branch3x3 = self.branch3x3_3(branch3x3)
# 因为kernel_size=(3, 3),为了保证图像宽高不改变,将padding设置为1,3/2=1
branch_pool = F.avg_pool2d(x, kernel_size=3, stride=1, padding=1)
branch_pool = self.branch_pool(branch_pool)
outputs = [branch1x1, branch5x5, branch3x3, branch_pool]
# (B, C, W, H),所以设置dim=1,即通过通道相连接起来
return torch.cat(outputs, dim=1)
class Net(torch.nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=(5, 5))
self.conv2 = torch.nn.Conv2d(88, 20, kernel_size=(5, 5)) # 16+24*3=88
self.incep1 = InceptionA(in_channels=10)
self.incep2 = InceptionA(in_channels=20)
self.mp = torch.nn.MaxPool2d(2)
self.fc = torch.nn.Linear(1408, 10)
def forward(self, x):
in_size = x.size(0)
x = F.relu(self.mp(self.conv1(x))) # 输入通道为1,输出通道为10
x = self.incep1(x) # 输入通道为10,输出通道为88
x = F.relu(self.mp(self.conv2(x))) # 输入通道为88,输出通道为20
x = self.incep2(x) # 输入通道为20,输出通道为88
x = x.view(in_size, -1)
x = self.fc(x)
return x
model = Net()
三、使用该Inception Module实现MNIST
分为四个步骤:
1.Prepare dataset
2.Design model using Class
3.Using PyTorch API to Construct loss and optimizer
4.Training Cycle(forward,backward,update)
完整代码如下:
# Implementation of Inception Module
import torch
from torchvision import transforms
from torchvision import datasets
from torch.utils.data import DataLoader
import torch.nn.functional as F
import torch.optim as optim
# -----------------------------------------------1.Prepare dataset------------------------------------------------------
batch_size = 64
# transforms.ToTensor():Covert the PIL Image to Tensor
# transforms.Normalize:The PARAMETERS are mean and std respectively,It use formulation x=(x-mean)/std
transform = transforms.Compose([transforms.ToTensor(),
transforms.Normalize((0.1307,), (0.3081,))])
train_dataset = datasets.MNIST(root='./dataset/mnist/',
train=True,
download=True,
transform=transform)
train_loader = DataLoader(train_dataset,
shuffle=True,
batch_size=batch_size)
test_dataset = datasets.MNIST(root='./dataset/mnist/',
train=False,
download=True,
transform=transform)
test_loader = DataLoader(test_dataset,
shuffle=False,
batch_size=batch_size)
# ----------------------------------------------------------------------------------------------------------------------
# -----------------------------------------2.Design model using Class---------------------------------------------------
class InceptionA(torch.nn.Module):
def __init__(self, in_channels):
super(InceptionA, self).__init__()
self.branch1x1 = torch.nn.Conv2d(in_channels, 16, kernel_size=(1, 1))
self.branch5x5_1 = torch.nn.Conv2d(in_channels, 16, kernel_size=(1, 1))
# 因为kernel_size=(5, 5),为了保证图像宽高不改变,将padding设置为2,5/2=2
self.branch5x5_2 = torch.nn.Conv2d(16, 24, kernel_size=(5, 5), padding=(2, 2))
self.branch3x3_1 = torch.nn.Conv2d(in_channels, 16, kernel_size=(1, 1))
# 因为kernel_size=(3, 3),为了保证图像宽高不改变,将padding设置为1,3/2=1
self.branch3x3_2 = torch.nn.Conv2d(16, 24, kernel_size=(3, 3), padding=(1, 1))
# 因为kernel_size=(3, 3),为了保证图像宽高不改变,将padding设置为1,3/2=1
self.branch3x3_3 = torch.nn.Conv2d(24, 24, kernel_size=(3, 3), padding=(1, 1))
self.branch_pool = torch.nn.Conv2d(in_channels, 24, kernel_size=(1, 1))
def forward(self, x):
branch1x1 = self.branch1x1(x)
branch5x5 = self.branch5x5_1(x)
branch5x5 = self.branch5x5_2(branch5x5)
branch3x3 = self.branch3x3_1(x)
branch3x3 = self.branch3x3_2(branch3x3)
branch3x3 = self.branch3x3_3(branch3x3)
# 因为kernel_size=(3, 3),为了保证图像宽高不改变,将padding设置为1,3/2=1
branch_pool = F.avg_pool2d(x, kernel_size=3, stride=1, padding=1)
branch_pool = self.branch_pool(branch_pool)
outputs = [branch1x1, branch5x5, branch3x3, branch_pool]
# (B, C, W, H),所以设置dim=1,即通过通道相连接起来
return torch.cat(outputs, dim=1)
class Net(torch.nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=(5, 5))
self.conv2 = torch.nn.Conv2d(88, 20, kernel_size=(5, 5)) # 16+24*3=88
self.incep1 = InceptionA(in_channels=10)
self.incep2 = InceptionA(in_channels=20)
self.mp = torch.nn.MaxPool2d(2)
self.fc = torch.nn.Linear(1408, 10)
def forward(self, x):
in_size = x.size(0)
x = F.relu(self.mp(self.conv1(x))) # 输入通道为1,输出通道为10
x = self.incep1(x) # 输入通道为10,输出通道为88
x = F.relu(self.mp(self.conv2(x))) # 输入通道为88,输出通道为20
x = self.incep2(x) # 输入通道为20,输出通道为88
x = x.view(in_size, -1)
x = self.fc(x)
return x
model = Net()
# ----------------------------------------------------------------------------------------------------------------------
# ---------------------------------3.Using PyTorch API to Construct loss and optimizer----------------------------------
criterion = torch.nn.CrossEntropyLoss()
# momentum是冲量,可以从局部极值走出来尽可能找到全局最优解
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.5)
# ----------------------------------------------------------------------------------------------------------------------
# ---------------------------------------------4.Training Cycle(forward,backward,update)--------------------------------
def train(epoch):
running_loss = 0.0
for batch_idx, data in enumerate(train_loader, 0):
inputs, target = data
# print(target)
optimizer.zero_grad()
# forward + backward + update
outputs = model(inputs)
# print(outputs)
loss = criterion(outputs, target)
loss.backward()
optimizer.step()
running_loss += loss.item()
if batch_idx % 300 == 299:
print('[%d, %5d] loss: %.3f' % (epoch + 1, batch_idx + 1, running_loss / 300))
running_loss = 0.0
def test():
correct = 0
total = 0
with torch.no_grad():
for data in test_loader:
images, labels = data
outputs = model(images)
# print(outputs)
_, predicted = torch.max(outputs.data, dim=1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
print('Accuracy on test set: %d %%' % (100 * correct / total))
# ----------------------------------------------------------------------------------------------------------------------
if __name__ == '__main__':
for epoch in range(10):
train(epoch)
test()
原视频出处https://www.bilibili.com/video/BV1Y7411d7Ys