第十讲卷积神经网络(基础篇)笔记与源码
B站 刘二大人 教学视频 传送门:卷积神经网络(基础篇)
视频截图分析:
1、首先卷积神经网络的过程大概是如下这个图的形式,首先将图片pillow转换为模型能够训练的张量模式,进行卷积特征提取,最后进行分类。
这里将拿到的图片pillow转换为张量的代码如下:
transform = transforms.Compose([transforms.ToTensor()])
2、对于卷积的输入和输出是怎么样的,进行3x3的矩阵的点乘,形成的这个位置的值就是他们所有位置点乘的值的和,这里使用单输入通道做例子如下图所示:(注:中间是卷积核)
第二个位置则是将3X3矩阵平移一个单位得出的结果,如此同理可得出ouput矩阵
所以很明显在3通道的情况下 可以得出3个矩阵,那么我们将其相加即可获得最后卷积输出矩阵。如下图所示:
也就是这样的:
如图所示:在m个filters时:
3、Padding
这里的意思其实就是比如 输入是5X5 但是卷积核是等于通道3的,那么他们卷积后会形成3X3的output,那么我们可以在外面加一圈0,那么生的output也是5X5的,这是为了实现想让图像大小不变。这里如何取判断加几圈,有个简单的方式是:比如卷积核=3,那么就是3/2=1,如果是5 那么5/2=2.如下图所示:
4、Stride=2
它每次框格移动的时候计算卷积是移动了两个格子,最后计算出的output会减小一半,如下图所示:
5、最大池化层
如用2x2Maxpooling 即画2X2矩阵 选出其中最大的那个值放入输出,然后Stride=2,开始计算,如果用2x2Maxpooling 那么最后获的得输出会小一半。
6、现在我们要实现的是一个这样的过程的卷积神经网络
卷积5x5-->2X2池化-->卷积5X5-->2X2池化-->线性层-->分类器
具体纵向过程如下:
代码说明:首先是加载数据集,pytroch提供了一个数据集,即使没有下载数据集,在相应你给的地址也会自动下载,贴心pytorch,哈哈哈哈!
代码源码:(注:这里用了GPU,若用CPU可把三行代码去掉或者直接用)
import torch
import matplotlib.pyplot as plt
#For constructing DataLoader
from torchvision import transforms,datasets
from torch.utils.data import DataLoader
#For using function relu()
import torch.nn.functional as F
#For constructing optimizer
import torch.optim as optim
import os
os.environ["KMP_DUPLICATE_LIB_OK"]='True'
batch_size = 64
transform = transforms.Compose([
#Convert the PLI Image to Tensor
transforms.ToTensor(),
#参数为平均值和标准值
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
)
class Net(torch.nn.Module):
def __init__(self):
super(Net,self).__init__()
self.conv1=torch.nn.Conv2d(1,10,kernel_size=5)
self.conv2=torch.nn.Conv2d(10,20,kernel_size=5)
self.pooling=torch.nn.MaxPool2d(2)
self.fc=torch.nn.Linear(320,10)
def forward(self,x):
#Flatten data from (n,1,28,28) to (n,784)
batch_size=x.size(0)
x=F.relu(self.pooling(self.conv1(x)))
x=F.relu(self.pooling(self.conv2(x)))
#为了给全连接的线性层用,我们把二位图片,20x4x4的变成编程一维的
x=x.view(batch_size,-1) # flattern -1 此处自动算出是320
x=self.fc(x)
return x
model = Net()
#Convert parameters and buffers of all modules to CUDA Tensor.
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model.to(device)
#多分类CrossEntropyLoss 二分类则用BCELoss
criterion=torch.nn.CrossEntropyLoss()
optimizer=optim.SGD(model.parameters(),lr=0.01,momentum=0.5)
def train(epoch):
running_loss = 0.0
for batch_idx, data in enumerate(train_loader, 0):
inputs, target = data
#Send the inputs and targets at every step to the GPU.
inputs, target = inputs.to(device), target.to(device)
optimizer.zero_grad()
# forward + backward + update
outputs = model(inputs)
loss = criterion(outputs, target)
#反向求梯度
loss.backward()
#更新权重
optimizer.step()
#让损失加起来
running_loss += loss.item()
if batch_idx % 300 == 299:
print('[%d, %5d] loss: %.3f' % (epoch + 1, batch_idx + 1, running_loss / 2000))
running_loss = 0.
epoch_list=[]
accept_list=[]
def test2():
correct = 0
total = 0
with torch.no_grad():#不用计算梯度
for data in test_loader:
images, target = data
#Send the inputs and targets at every step to the GPU.
images, target = images.to(device), target.to(device)
outputs = model(images)
#我们取概率最大的那个数输出
_, predicted = torch.max(outputs.data, dim=1)
total += target.size(0)
#计算正确率
correct += (predicted == target).sum().item()
accept_list.append(100 * correct / total)
print('Accuracy on test set: %d %% [%d/%d]' % (100 * correct / total, correct, total))
if __name__=='__main__':
for epoch in range(10):
train(epoch)
test2()
epoch_list.append(epoch)
plt.plot(epoch_list,accept_list)
plt.xlabel("epoch")
plt.rcParams['font.sans-serif'] = ['SimHei'] # 设置中文字体
plt.ylabel("成功率 单位为 %")
plt.show()
输出结果:
Files already downloaded
Files already downloaded
[1, 300] loss: 0.094
[1, 600] loss: 0.028
[1, 900] loss: 0.021
Accuracy on test set: 96 % [9673/10000]
[2, 300] loss: 0.017
[2, 600] loss: 0.015
[2, 900] loss: 0.014
Accuracy on test set: 97 % [9766/10000]
[3, 300] loss: 0.013
[3, 600] loss: 0.012
[3, 900] loss: 0.010
Accuracy on test set: 98 % [9835/10000]
[4, 300] loss: 0.010
[4, 600] loss: 0.009
[4, 900] loss: 0.009
Accuracy on test set: 98 % [9828/10000]
[5, 300] loss: 0.008
[5, 600] loss: 0.008
[5, 900] loss: 0.009
Accuracy on test set: 98 % [9835/10000]
[6, 300] loss: 0.008
[6, 600] loss: 0.007
[6, 900] loss: 0.008
Accuracy on test set: 98 % [9868/10000]
[7, 300] loss: 0.007
[7, 600] loss: 0.007
[7, 900] loss: 0.007
Accuracy on test set: 98 % [9866/10000]
[8, 300] loss: 0.006
[8, 600] loss: 0.007
[8, 900] loss: 0.006
Accuracy on test set: 98 % [9879/10000]
[9, 300] loss: 0.006
[9, 600] loss: 0.005
[9, 900] loss: 0.006
Accuracy on test set: 98 % [9858/10000]
[10, 300] loss: 0.006
[10, 600] loss: 0.005
[10, 900] loss: 0.006
Accuracy on test set: 98 % [9882/10000]
运行效果图:
第二部分:刘老师给的作业内容,给出代码和过程
本人学生 如有错误敬请指正。谢谢,(作业内容)
手算的特征数图:
代码:(主要改的是模型部份,所以代码只贴模型部份)
class Net(torch.nn.Module):
def __init__(self):
super(Net,self).__init__()
self.conv1=torch.nn.Conv2d(1,10,kernel_size=5)
self.conv2=torch.nn.Conv2d(10,15,kernel_size=5,padding=2)
self.conv3=torch.nn.Conv2d(15,20,5)
self.pooling=torch.nn.MaxPool2d(2)
self.fc1=torch.nn.Linear(20,15)
self.fc2=torch.nn.Linear(15,10)
def forward(self,x):
batch_size=x.size(0)
x=F.relu(self.conv1(x))
x=F.relu(self.conv2(self.pooling(x)))
x=F.relu(self.conv3(self.pooling(x)))
x=F.relu(self.pooling(x))
x=x.view(batch_size,-1)
x=F.relu(self.fc1(x))
return self.fc2(x)
运行结果:
[1, 300] loss: 0.305
[1, 600] loss: 0.084
[1, 900] loss: 0.042
Accuracy on test set: 92 % [9289/10000]
[2, 300] loss: 0.029
[2, 600] loss: 0.023
[2, 900] loss: 0.019
Accuracy on test set: 95 % [9590/10000]
[3, 300] loss: 0.016
[3, 600] loss: 0.016
[3, 900] loss: 0.014
Accuracy on test set: 97 % [9758/10000]
[4, 300] loss: 0.013
[4, 600] loss: 0.011
[4, 900] loss: 0.012
Accuracy on test set: 98 % [9800/10000]
[5, 300] loss: 0.010
[5, 600] loss: 0.009
[5, 900] loss: 0.010
Accuracy on test set: 97 % [9778/10000]
[6, 300] loss: 0.009
[6, 600] loss: 0.009
[6, 900] loss: 0.008
Accuracy on test set: 98 % [9829/10000]
[7, 300] loss: 0.007
[7, 600] loss: 0.007
[7, 900] loss: 0.008
Accuracy on test set: 98 % [9849/10000]
[8, 300] loss: 0.007
[8, 600] loss: 0.006
[8, 900] loss: 0.007
Accuracy on test set: 98 % [9805/10000]
[9, 300] loss: 0.006
[9, 600] loss: 0.006
[9, 900] loss: 0.006
Accuracy on test set: 98 % [9859/10000]
[10, 300] loss: 0.005
[10, 600] loss: 0.006
[10, 900] loss: 0.006
Accuracy on test set: 98 % [9875/10000]
进程已结束,退出代码为 0
运行效果图:
完结,若有任何错误,敬请指正,万分感谢!!