卷积网络模型CNN的实现
结构以及计算
CNN主要包括以下结构：
输入层(Input layer)：输入数据；
卷积层(Convolution layer，CONV)：使用卷积核进行特征提取和特征映射；
激活层：非线性映射（ReLU)
池化层(Pooling layer，POOL)：进行下采样降维；
光栅化(Rasterization)：展开像素，与全连接层全连接，某些情况下这一层可以省去；
全连接层(Affine layer / Fully Connected layer，FC)：在尾部进行拟合，减少特征信息的损失；
激活层：非线性映射（ReLU)
输出层(Output layer)：输出结果。
卷积后shape计算参考我写的：卷积神经网络的计算
| input | process | outpout | |
|---|---|---|---|
| 卷积层 | 1 ∗ 28 ∗ 28 1* 28* 28 1∗28∗28 | in_channels=1,out_channels=16 kernel_size=5,stride=1,padding=2 输出长度= 28 + 2 ∗ 2 − 5 + 1 = 28 28+2*2-5+1=28 28+2∗2−5+1=28 |
16 ∗ 28 ∗ 28 16*28*28 16∗28∗28 |
| ReLU | 16 ∗ 28 ∗ 28 16*28*28 16∗28∗28 | 16 ∗ 28 ∗ 28 16*28*28 16∗28∗28 | |
| MaxPool2d | 16 ∗ 28 ∗ 28 16*28*28 16∗28∗28 | in_channels=16,out_channels=16 kernel_size=2,strides=2,padding=0 输出长度= ( 28 − 2 + 1 ) / 2 + 1 = 14 (28-2+1)/2+1=14 (28−2+1)/2+1=14 |
16 ∗ 14 ∗ 14 16*14*14 16∗14∗14 |
| 卷积层 | 16 ∗ 14 ∗ 14 16*14*14 16∗14∗14 | in_channels=16, out_channels=32 kernel_size=5, stride=1, padding=2 输出长度= 14 + 2 ∗ 2 − 5 + 1 = 14 14+2*2-5+1=14 14+2∗2−5+1=14 |
32 ∗ 14 ∗ 14 32*14*14 32∗14∗14 |
| ReLU | 32 ∗ 14 ∗ 14 32*14*14 32∗14∗14 | 32 ∗ 14 ∗ 14 32*14*14 32∗14∗14 | |
| MaxPool2d | 32 ∗ 14 ∗ 14 32*14*14 32∗14∗14 | in_channels=32,out_channels=32 kernel_size=2,strides=2,padding=0 输出长度=$(14-2+1)/2+1=7 |
32 ∗ 7 ∗ 7 32*7*7 32∗7∗7 |
| 全连接层 | 32 ∗ 7 ∗ 7 32*7*7 32∗7∗7 | 32 ∗ 7 ∗ 7 32*7*7 32∗7∗7 |
代码
CNN
import torch
from torch import nn
class CNN(nn.Module):
def __init__(self):
super(CNN, self).__init__()
self.conv1 = nn.Sequential( # input shape (1, 28, 28)
nn.Conv2d(
in_channels=1, # input height gray just have one level
out_channels=16, # n_filters
kernel_size=5, # filter size
stride=1, # filter movement/step
padding=2
), # output shape (16, 28, 28)
#nn.ReLU(), # activation
nn.MaxPool2d(kernel_size=2,stride=2,padding=0), # choose max value in 2x2 area, output shape (16, 14, 14)
)
self.conv2 = nn.Sequential( # input shape (16, 14, 14)
nn.Conv2d(16,32, 5, 1, 2), # output shape (32, 14, 14)
#nn.ReLU(), # activation
nn.MaxPool2d(2,stride=2), # output shape (32, 7, 7)
)
# flatten
# fc unit
self.fc1 = nn.Sequential(
nn.Linear(32*7*7,500),
#nn.ReLU()
)
self.fc2 = nn.Sequential(
nn.Linear(500,10),
#nn.ReLU()
)
def forward(self, x):
batchsz = x.size(0)
x = self.conv1(x)
x = self.conv2(x)
x = x.view(batchsz, 50*4*4)
x= self.fc1(x)
logits = self.fc2(x)
# # [b, 10]
# pred = F.softmax(logits, dim=1)
# loss = self.criteon(logits, y)
return logits
from torchstat import stat
from CNN import CNN
# 导入模型，输入一张输入图片的尺寸
cnn=CNN()
print(cnn)
stat(cnn,(1, 28,28))
