1、LeNet 模型
LeNet分为卷积层块和全连接层块两个部分。下面我们分别介绍这两个模块。
（1）卷积层块里的基本单位是卷积层后接平均池化层：卷积层用来识别图像里的空间模式，如线条和物体局部，之后的平均池化层则用来降低卷积层对位置的敏感性。
（2）卷积层块由两个这样的基本单位重复堆叠构成。在卷积层块中，每个卷积层都使用5×55×5 的窗口，并在输出上使用sigmoid激活函数。第一个卷积层输出通道数为6，第二个卷积层输出通道数则增加到16。
（3）全连接层块含3个全连接层。它们的输出个数分别是120、84和10，其中10为输出的类别个数。
下面我们通过Sequential类来实现LeNet模型。
In [1]:
#import
import sys
sys.path.append("/home/kesci/input")
import d2lzh1981 as d2l
import torch
import torch.nn as nn
import torch.optim as optim
import time
In [2]:
#net
class Flatten(torch.nn.Module): #展平操作
def forward(self, x):
return x.view(x.shape[0], -1)
class Reshape(torch.nn.Module): #将图像大小重定型
def forward(self, x):
return x.view(-1,1,28,28) #(B x C x H x W)
net = torch.nn.Sequential( #Lelet
Reshape(),
nn.Conv2d(in_channels=1, out_channels=6, kernel_size=5, padding=2), #b*1*28*28 =>b*6*28*28
nn.Sigmoid(),
nn.AvgPool2d(kernel_size=2, stride=2), #b*6*28*28 =>b*6*14*14
nn.Conv2d(in_channels=6, out_channels=16, kernel_size=5), #b*6*14*14 =>b*16*10*10
nn.Sigmoid(),
nn.AvgPool2d(kernel_size=2, stride=2), #b*16*10*10 => b*16*5*5
Flatten(), #b*16*5*5 => b*400
nn.Linear(in_features=16*5*5, out_features=120),
nn.Sigmoid(),
nn.Linear(120, 84),
nn.Sigmoid(),
nn.Linear(84, 10)
)
接下来我们构造一个高和宽均为28的单通道数据样本，并逐层进行前向计算来查看每个层的输出形状。
In [3]:
#print
X = torch.randn(size=(1,1,28,28), dtype = torch.float32)
for layer in net:
X = layer(X)
print(layer.__class__.__name__,'output shape: \t',X.shape)
Reshape output shape: torch.Size([1, 1, 28, 28])
Conv2d output shape: torch.Size([1, 6, 28, 28])
Sigmoid output shape: torch.Size([1, 6, 28, 28])
AvgPool2d output shape: torch.Size([1, 6, 14, 14])
Conv2d output shape: torch.Size([1, 16, 10, 10])
Sigmoid output shape: torch.Size([1, 16, 10, 10])
AvgPool2d output shape: torch.Size([1, 16, 5, 5])
Flatten output shape: torch.Size([1, 400])
Linear output shape: torch.Size([1, 120])
Sigmoid output shape: torch.Size([1, 120])
Linear output shape: torch.Size([1, 84])
Sigmoid output shape: torch.Size([1, 84])
Linear output shape: torch.Size([1, 10])
可以看到，在卷积层块中输入的高和宽在逐层减小。卷积层由于使用高和宽均为5的卷积核，从而将高和宽分别减小4，而池化层则将高和宽减半，但通道数则从1增加到16。全连接层则逐层减少输出个数，直到变成图像的类别数10。
获取数据和训练模型¶
下面我们来实现LeNet模型。我们使用Fashion-MNIST作为训练数据集。
In [4]:
# 数据
batch_size = 256
train_iter, test_iter = d2l.load_data_fashion_mnist(
batch_size=batch_size, root='/home/kesci/input/FashionMNIST2065')
print(len(train_iter))
235
为了使读者更加形象的看到数据，添加额外的部分来展示数据的图像
In [5]:
#数据展示
import matplotlib.pyplot as plt
def show_fashion_mnist(images, labels):
d2l.use_svg_display()
# 这里的_表示我们忽略（不使用）的变量
_, figs = plt.subplots(1, len(images), figsize=(12, 12))
for f, img, lbl in zip(figs, images, labels):
f.imshow(img.view((28, 28)).numpy())
f.set_title(lbl)
f.axes.get_xaxis().set_visible(False)
f.axes.get_yaxis().set_visible(False)
plt.show()
for Xdata,ylabel in train_iter:
break
X, y = [], []
for i in range(10):
print(Xdata[i].shape,ylabel[i].numpy())
X.append(Xdata[i]) # 将第i个feature加到X中
y.append(ylabel[i].numpy()) # 将第i个label加到y中
show_fashion_mnist(X, y)
torch.Size([1, 28, 28]) 3
torch.Size([1, 28, 28]) 8
torch.Size([1, 28, 28]) 1
torch.Size([1, 28, 28]) 4
torch.Size([1, 28, 28]) 0
torch.Size([1, 28, 28]) 0
torch.Size([1, 28, 28]) 4
torch.Size([1, 28, 28]) 9
torch.Size([1, 28, 28]) 4
torch.Size([1, 28, 28]) 7
因为卷积神经网络计算比多层感知机要复杂，建议使用GPU来加速计算。我们查看看是否可以用GPU，如果成功则使用cuda:0，否则仍然使用cpu。
In [6]:
# This function has been saved in the d2l package for future use
#use GPU
def try_gpu():
"""If GPU is available, return torch.device as cuda:0; else return torch.device as cpu."""
if torch.cuda.is_available():
device = torch.device('cuda:0')
else:
device = torch.device('cpu')
return device
device = try_gpu()
device
Out[6]:
device(type='cpu')
我们实现evaluate_accuracy函数，该函数用于计算模型net在数据集data_iter上的准确率。
In [7]:
#计算准确率
'''
(1). net.train()
启用 BatchNormalization 和 Dropout，将BatchNormalization和Dropout置为True
(2). net.eval()
不启用 BatchNormalization 和 Dropout，将BatchNormalization和Dropout置为False
'''
def evaluate_accuracy(data_iter, net,device=torch.device('cpu')):
"""Evaluate accuracy of a model on the given data set."""
acc_sum,n = torch.tensor([0],dtype=torch.float32,device=device),0
for X,y in data_iter:
# If device is the GPU, copy the data to the GPU.
X,y = X.to(device),y.to(device)
net.eval()
with torch.no_grad():
y = y.long()
acc_sum += torch.sum((torch.argmax(net(X), dim=1) == y)) #[[0.2 ,0.4 ,0.5 ,0.6 ,0.8] ,[ 0.1,0.2 ,0.4 ,0.3 ,0.1]] => [ 4 , 2 ]
n += y.shape[0]
return acc_sum.item()/n
我们定义函数train_ch5，用于训练模型。
In [8]:
#训练函数
def train_ch5(net, train_iter, test_iter,criterion, num_epochs, batch_size, device,lr=None):
"""Train and evaluate a model with CPU or GPU."""
print('training on', device)
net.to(device)
optimizer = optim.SGD(net.parameters(), lr=lr)
for epoch in range(num_epochs):
train_l_sum = torch.tensor([0.0],dtype=torch.float32,device=device)
train_acc_sum = torch.tensor([0.0],dtype=torch.float32,device=device)
n, start = 0, time.time()
for X, y in train_iter:
net.train()
optimizer.zero_grad()
X,y = X.to(device),y.to(device)
y_hat = net(X)
loss = criterion(y_hat, y)
loss.backward()
optimizer.step()
with torch.no_grad():
y = y.long()
train_l_sum += loss.float()
train_acc_sum += (torch.sum((torch.argmax(y_hat, dim=1) == y))).float()
n += y.shape[0]
test_acc = evaluate_accuracy(test_iter, net,device)
print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f, '
'time %.1f sec'
% (epoch + 1, train_l_sum/n, train_acc_sum/n, test_acc,
time.time() - start))
我们重新将模型参数初始化到对应的设备device(cpu or cuda:0)之上，并使用Xavier随机初始化。损失函数和训练算法则依然使用交叉熵损失函数和小批量随机梯度下降。
In [9]:
# 训练
lr, num_epochs = 0.9, 10
def init_weights(m):
if type(m) == nn.Linear or type(m) == nn.Conv2d:
torch.nn.init.xavier_uniform_(m.weight)
net.apply(init_weights)
net = net.to(device)
criterion = nn.CrossEntropyLoss() #交叉熵描述了两个概率分布之间的距离，交叉熵越小说明两者之间越接近
train_ch5(net, train_iter, test_iter, criterion,num_epochs, batch_size,device, lr)
training on cpu
epoch 1, loss 0.0091, train acc 0.100, test acc 0.168, time 21.6 sec
epoch 2, loss 0.0065, train acc 0.355, test acc 0.599, time 21.5 sec
epoch 3, loss 0.0035, train acc 0.651, test acc 0.665, time 21.8 sec
epoch 4, loss 0.0028, train acc 0.717, test acc 0.723, time 21.7 sec
epoch 5, loss 0.0025, train acc 0.746, test acc 0.753, time 21.4 sec
epoch 6, loss 0.0023, train acc 0.767, test acc 0.754, time 21.5 sec
epoch 7, loss 0.0022, train acc 0.782, test acc 0.785, time 21.3 sec
epoch 8, loss 0.0021, train acc 0.798, test acc 0.791, time 21.8 sec
epoch 9, loss 0.0019, train acc 0.811, test acc 0.790, time 22.0 sec
epoch 10, loss 0.0019, train acc 0.821, test acc 0.804, time 22.1 sec
In [10]:
# test
for testdata,testlabe in test_iter:
testdata,testlabe = testdata.to(device),testlabe.to(device)
break
print(testdata.shape,testlabe.shape)
net.eval()
y_pre = net(testdata)
print(torch.argmax(y_pre,dim=1)[:10])
print(testlabe[:10])
torch.Size([256, 1, 28, 28]) torch.Size([256])
tensor([9, 2, 1, 1, 6, 1, 2, 6, 5, 7])
tensor([9, 2, 1, 1, 6, 1, 4, 6, 5, 7])
总结：¶
卷积神经网络就是含卷积层的网络。 LeNet交替使用卷积层和最大池化层后接全连接层来进行图像分类。
2、AlexNet网络
LeNet: 在大的真实数据集上的表现并不尽如⼈意。
1.神经网络计算复杂。
2.还没有⼤量深⼊研究参数初始化和⾮凸优化算法等诸多领域。
AlexNet¶
首次证明了学习到的特征可以超越⼿⼯设计的特征，从而⼀举打破计算机视觉研究的前状。
特征：
- 8层变换，其中有5层卷积和2层全连接隐藏层，以及1个全连接输出层。
- 将sigmoid激活函数改成了更加简单的ReLU激活函数。
- 用Dropout来控制全连接层的模型复杂度。
- 引入数据增强，如翻转、裁剪和颜色变化，从而进一步扩大数据集来缓解过拟合。
In [1]:
import time
import torch
from torch import nn, optim
import torchvision
import numpy as np
import sys
sys.path.append("/home/kesci/input/")
import d2lzh1981 as d2l
import os
import torch.nn.functional as F
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
class AlexNet(nn.Module):
def __init__(self):
super(AlexNet, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(1, 96, 11, 4), # in_channels, out_channels, kernel_size, stride, padding
nn.ReLU(),
nn.MaxPool2d(3, 2), # kernel_size, stride
# 减小卷积窗口，使用填充为2来使得输入与输出的高和宽一致，且增大输出通道数
nn.Conv2d(96, 256, 5, 1, 2),
nn.ReLU(),
nn.MaxPool2d(3, 2),
# 连续3个卷积层，且使用更小的卷积窗口。除了最后的卷积层外，进一步增大了输出通道数。
# 前两个卷积层后不使用池化层来减小输入的高和宽
nn.Conv2d(256, 384, 3, 1, 1),
nn.ReLU(),
nn.Conv2d(384, 384, 3, 1, 1),
nn.ReLU(),
nn.Conv2d(384, 256, 3, 1, 1),
nn.ReLU(),
nn.MaxPool2d(3, 2)
)
# 这里全连接层的输出个数比LeNet中的大数倍。使用丢弃层来缓解过拟合
self.fc = nn.Sequential(
nn.Linear(256*5*5, 4096),
nn.ReLU(),
nn.Dropout(0.5),
#由于使用CPU镜像，精简网络，若为GPU镜像可添加该层
#nn.Linear(4096, 4096),
#nn.ReLU(),
#nn.Dropout(0.5),
# 输出层。由于这里使用Fashion-MNIST，所以用类别数为10，而非论文中的1000
nn.Linear(4096, 10),
)
def forward(self, img):
feature = self.conv(img)
output = self.fc(feature.view(img.shape[0], -1))
return output
In [2]:
net = AlexNet()
print(net)
AlexNet(
(conv): Sequential(
(0): Conv2d(1, 96, kernel_size=(11, 11), stride=(4, 4))
(1): ReLU()
(2): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
(3): Conv2d(96, 256, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(4): ReLU()
(5): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
(6): Conv2d(256, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(7): ReLU()
(8): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(9): ReLU()
(10): Conv2d(384, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU()
(12): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(fc): Sequential(
(0): Linear(in_features=6400, out_features=4096, bias=True)
(1): ReLU()
(2): Dropout(p=0.5, inplace=False)
(3): Linear(in_features=4096, out_features=10, bias=True)
)
)
载入数据集¶
In [3]:
# 本函数已保存在d2lzh_pytorch包中方便以后使用
def load_data_fashion_mnist(batch_size, resize=None, root='/home/kesci/input/FashionMNIST2065'):
"""Download the fashion mnist dataset and then load into memory."""
trans = []
if resize:
trans.append(torchvision.transforms.Resize(size=resize))
trans.append(torchvision.transforms.ToTensor())
transform = torchvision.transforms.Compose(trans)
mnist_train = torchvision.datasets.FashionMNIST(root=root, train=True, download=True, transform=transform)
mnist_test = torchvision.datasets.FashionMNIST(root=root, train=False, download=True, transform=transform)
train_iter = torch.utils.data.DataLoader(mnist_train, batch_size=batch_size, shuffle=True, num_workers=2)
test_iter = torch.utils.data.DataLoader(mnist_test, batch_size=batch_size, shuffle=False, num_workers=2)
return train_iter, test_iter
#batchsize=128
batch_size = 16
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = load_data_fashion_mnist(batch_size,224)
for X, Y in train_iter:
print('X =', X.shape,
'\nY =', Y.type(torch.int32))
break
X = torch.Size([16, 1, 224, 224])
Y = tensor([5, 2, 9, 3, 1, 8, 3, 3, 2, 6, 1, 6, 2, 4, 4, 8], dtype=torch.int32)
训练¶
In [4]:
lr, num_epochs = 0.001, 3
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs)
3、VGG
VGG：通过重复使⽤简单的基础块来构建深度模型。
Block:数个相同的填充为1、窗口形状为3×33×3 的卷积层,接上一个步幅为2、窗口形状为2×22×2 的最大池化层。
卷积层保持输入的高和宽不变，而池化层则对其减半。
VGG11的简单实现¶
In [4]:
def vgg_block(num_convs, in_channels, out_channels): #卷积层个数，输入通道数，输出通道数
blk = []
for i in range(num_convs):
if i == 0:
blk.append(nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1))
else:
blk.append(nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1))
blk.append(nn.ReLU())
blk.append(nn.MaxPool2d(kernel_size=2, stride=2)) # 这里会使宽高减半
return nn.Sequential(*blk)
In [5]:
conv_arch = ((1, 1, 64), (1, 64, 128), (2, 128, 256), (2, 256, 512), (2, 512, 512))
# 经过5个vgg_block, 宽高会减半5次, 变成 224/32 = 7
fc_features = 512 * 7 * 7 # c * w * h
fc_hidden_units = 4096 # 任意
In [6]:
def vgg(conv_arch, fc_features, fc_hidden_units=4096):
net = nn.Sequential()
# 卷积层部分
for i, (num_convs, in_channels, out_channels) in enumerate(conv_arch):
# 每经过一个vgg_block都会使宽高减半
net.add_module("vgg_block_" + str(i+1), vgg_block(num_convs, in_channels, out_channels))
# 全连接层部分
net.add_module("fc", nn.Sequential(d2l.FlattenLayer(),
nn.Linear(fc_features, fc_hidden_units),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(fc_hidden_units, fc_hidden_units),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(fc_hidden_units, 10)
))
return net
In [7]:
net = vgg(conv_arch, fc_features, fc_hidden_units)
X = torch.rand(1, 1, 224, 224)
# named_children获取一级子模块及其名字(named_modules会返回所有子模块,包括子模块的子模块)
for name, blk in net.named_children():
X = blk(X)
print(name, 'output shape: ', X.shape)
vgg_block_1 output shape: torch.Size([1, 64, 112, 112])
vgg_block_2 output shape: torch.Size([1, 128, 56, 56])
vgg_block_3 output shape: torch.Size([1, 256, 28, 28])
vgg_block_4 output shape: torch.Size([1, 512, 14, 14])
vgg_block_5 output shape: torch.Size([1, 512, 7, 7])
fc output shape: torch.Size([1, 10])
In [8]:
ratio = 8
small_conv_arch = [(1, 1, 64//ratio), (1, 64//ratio, 128//ratio), (2, 128//ratio, 256//ratio),
(2, 256//ratio, 512//ratio), (2, 512//ratio, 512//ratio)]
net = vgg(small_conv_arch, fc_features // ratio, fc_hidden_units // ratio)
print(net)
Sequential(
(vgg_block_1): Sequential(
(0): Conv2d(1, 8, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(vgg_block_2): Sequential(
(0): Conv2d(8, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(vgg_block_3): Sequential(
(0): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(vgg_block_4): Sequential(
(0): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(vgg_block_5): Sequential(
(0): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU()
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU()
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(fc): Sequential(
(0): FlattenLayer()
(1): Linear(in_features=3136, out_features=512, bias=True)
(2): ReLU()
(3): Dropout(p=0.5, inplace=False)
(4): Linear(in_features=512, out_features=512, bias=True)
(5): ReLU()
(6): Dropout(p=0.5, inplace=False)
(7): Linear(in_features=512, out_features=10, bias=True)
)
)
In [10]:
batchsize=16
#batch_size = 64
# 如出现“out of memory”的报错信息，可减小batch_size或resize
# train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)
lr, num_epochs = 0.001, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs)
4、⽹络中的⽹络（NiN）
LeNet、AlexNet和VGG：先以由卷积层构成的模块充分抽取 空间特征，再以由全连接层构成的模块来输出分类结果。
NiN：串联多个由卷积层和“全连接”层构成的小⽹络来构建⼀个深层⽹络。
⽤了输出通道数等于标签类别数的NiN块，然后使⽤全局平均池化层对每个通道中所有元素求平均并直接⽤于分类。
1×1卷积核作用
1.放缩通道数：通过控制卷积核的数量达到通道数的放缩。
2.增加非线性。1×1卷积核的卷积过程相当于全连接层的计算过程，并且还加入了非线性激活函数，从而可以增加网络的非线性。
3.计算参数少
In [9]:
def nin_block(in_channels, out_channels, kernel_size, stride, padding):
blk = nn.Sequential(nn.Conv2d(in_channels, out_channels, kernel_size, stride, padding),
nn.ReLU(),
nn.Conv2d(out_channels, out_channels, kernel_size=1),
nn.ReLU(),
nn.Conv2d(out_channels, out_channels, kernel_size=1),
nn.ReLU())
return blk
In [10]:
# 已保存在d2lzh_pytorch
class GlobalAvgPool2d(nn.Module):
# 全局平均池化层可通过将池化窗口形状设置成输入的高和宽实现
def __init__(self):
super(GlobalAvgPool2d, self).__init__()
def forward(self, x):
return F.avg_pool2d(x, kernel_size=x.size()[2:])
net = nn.Sequential(
nin_block(1, 96, kernel_size=11, stride=4, padding=0),
nn.MaxPool2d(kernel_size=3, stride=2),
nin_block(96, 256, kernel_size=5, stride=1, padding=2),
nn.MaxPool2d(kernel_size=3, stride=2),
nin_block(256, 384, kernel_size=3, stride=1, padding=1),
nn.MaxPool2d(kernel_size=3, stride=2),
nn.Dropout(0.5),
# 标签类别数是10
nin_block(384, 10, kernel_size=3, stride=1, padding=1),
GlobalAvgPool2d(),
# 将四维的输出转成二维的输出，其形状为(批量大小, 10)
d2l.FlattenLayer())
In [11]:
X = torch.rand(1, 1, 224, 224)
for name, blk in net.named_children():
X = blk(X)
print(name, 'output shape: ', X.shape)
0 output shape: torch.Size([1, 96, 54, 54])
1 output shape: torch.Size([1, 96, 26, 26])
2 output shape: torch.Size([1, 256, 26, 26])
3 output shape: torch.Size([1, 256, 12, 12])
4 output shape: torch.Size([1, 384, 12, 12])
5 output shape: torch.Size([1, 384, 5, 5])
6 output shape: torch.Size([1, 384, 5, 5])
7 output shape: torch.Size([1, 10, 5, 5])
8 output shape: torch.Size([1, 10, 1, 1])
9 output shape: torch.Size([1, 10])
In [14]:
batch_size = 128
# 如出现“out of memory”的报错信息，可减小batch_size或resize
#train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)
lr, num_epochs = 0.002, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs)
NiN重复使⽤由卷积层和代替全连接层的1×1卷积层构成的NiN块来构建深层⽹络。
NiN去除了容易造成过拟合的全连接输出层，而是将其替换成输出通道数等于标签类别数 的NiN块和全局平均池化层。
NiN的以上设计思想影响了后⾯⼀系列卷积神经⽹络的设计。
5、GoogLeNet
- 由Inception基础块组成。
- Inception块相当于⼀个有4条线路的⼦⽹络。它通过不同窗口形状的卷积层和最⼤池化层来并⾏抽取信息，并使⽤1×1卷积层减少通道数从而降低模型复杂度。
- 可以⾃定义的超参数是每个层的输出通道数，我们以此来控制模型复杂度。
In [12]:
class Inception(nn.Module):
# c1 - c4为每条线路里的层的输出通道数
def __init__(self, in_c, c1, c2, c3, c4):
super(Inception, self).__init__()
# 线路1，单1 x 1卷积层
self.p1_1 = nn.Conv2d(in_c, c1, kernel_size=1)
# 线路2，1 x 1卷积层后接3 x 3卷积层
self.p2_1 = nn.Conv2d(in_c, c2[0], kernel_size=1)
self.p2_2 = nn.Conv2d(c2[0], c2[1], kernel_size=3, padding=1)
# 线路3，1 x 1卷积层后接5 x 5卷积层
self.p3_1 = nn.Conv2d(in_c, c3[0], kernel_size=1)
self.p3_2 = nn.Conv2d(c3[0], c3[1], kernel_size=5, padding=2)
# 线路4，3 x 3最大池化层后接1 x 1卷积层
self.p4_1 = nn.MaxPool2d(kernel_size=3, stride=1, padding=1)
self.p4_2 = nn.Conv2d(in_c, c4, kernel_size=1)
def forward(self, x):
p1 = F.relu(self.p1_1(x))
p2 = F.relu(self.p2_2(F.relu(self.p2_1(x))))
p3 = F.relu(self.p3_2(F.relu(self.p3_1(x))))
p4 = F.relu(self.p4_2(self.p4_1(x)))
return torch.cat((p1, p2, p3, p4), dim=1) # 在通道维上连结输出
GoogLeNet模型¶
完整模型结构
In [16]:
b1 = nn.Sequential(nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3),
nn.ReLU(),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
b2 = nn.Sequential(nn.Conv2d(64, 64, kernel_size=1),
nn.Conv2d(64, 192, kernel_size=3, padding=1),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
b3 = nn.Sequential(Inception(192, 64, (96, 128), (16, 32), 32),
Inception(256, 128, (128, 192), (32, 96), 64),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
b4 = nn.Sequential(Inception(480, 192, (96, 208), (16, 48), 64),
Inception(512, 160, (112, 224), (24, 64), 64),
Inception(512, 128, (128, 256), (24, 64), 64),
Inception(512, 112, (144, 288), (32, 64), 64),
Inception(528, 256, (160, 320), (32, 128), 128),
nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
b5 = nn.Sequential(Inception(832, 256, (160, 320), (32, 128), 128),
Inception(832, 384, (192, 384), (48, 128), 128),
d2l.GlobalAvgPool2d())
net = nn.Sequential(b1, b2, b3, b4, b5,
d2l.FlattenLayer(), nn.Linear(1024, 10))
net = nn.Sequential(b1, b2, b3, b4, b5, d2l.FlattenLayer(), nn.Linear(1024, 10))
X = torch.rand(1, 1, 96, 96)
for blk in net.children():
X = blk(X)
print('output shape: ', X.shape)
#batchsize=128
batch_size = 16
# 如出现“out of memory”的报错信息，可减小batch_size或resize
#train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=96)
lr, num_epochs = 0.001, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
d2l.train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs)
800
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
