VGG使用可以重复使用的卷积快来构建深度卷积神经网络
不同的卷积块个数和超参数可以得到不同复杂程度的变种
VGG对空间特征提取好一些,但是对配置要求更高,同样是因为我电脑显存的问题无法读取
Python 3.8.8 (default, Apr 13 2021, 15:08:03) [MSC v.1916 64 bit (AMD64)]
Type 'copyright', 'credits' or 'license' for more information
IPython 7.22.0 -- An enhanced Interactive Python. Type '?' for help.
PyDev console: using IPython 7.22.0
Python 3.8.8 (default, Apr 13 2021, 15:08:03) [MSC v.1916 64 bit (AMD64)] on win32
import torch
from torch import nn
from d2l import torch as d2l
def vgg_block(num_convs, in_channels, out_channels): #多少个卷积层 输入数 输出书
layers = []
for _ in range(num_convs):
layers.append(nn.Conv2d(in_channels, out_channels,
kernel_size=3, padding=1))
layers.append(nn.ReLU())
in_channels = out_channels
layers.append(nn.MaxPool2d(kernel_size=2,stride=2))
return nn.Sequential(*layers)
conv_arch = ((1, 6), (1, 12), (2, 24), (2, 48), (2, 96)) #五块
def vgg(conv_arch):
conv_blks = []
in_channels = 1
# 卷积层部分
for (num_convs, out_channels) in conv_arch:
conv_blks.append(vgg_block(num_convs, in_channels, out_channels))
in_channels = out_channels
return nn.Sequential(
*conv_blks, nn.Flatten(),
# 全连接层部分
nn.Linear(out_channels * 7 * 7, 768), nn.ReLU(), nn.Dropout(0.5),
nn.Linear(768, 768), nn.ReLU(), nn.Dropout(0.5),
nn.Linear(768, 10))
net = vgg(conv_arch)
X = torch.randn(size=(1, 1, 224, 224))
for blk in net:
X = blk(X)
print(blk.__class__.__name__,'output shape:\t',X.shape)
Sequential output shape: torch.Size([1, 6, 112, 112])
Sequential output shape: torch.Size([1, 12, 56, 56])
Sequential output shape: torch.Size([1, 24, 28, 28])
Sequential output shape: torch.Size([1, 48, 14, 14])
Sequential output shape: torch.Size([1, 96, 7, 7])
Flatten output shape: torch.Size([1, 4704])
Linear output shape: torch.Size([1, 768])
ReLU output shape: torch.Size([1, 768])
Dropout output shape: torch.Size([1, 768])
Linear output shape: torch.Size([1, 768])
ReLU output shape: torch.Size([1, 768])
Dropout output shape: torch.Size([1, 768])
Linear output shape: torch.Size([1, 10])
ratio = 4
small_conv_arch = [(pair[0], pair[1] // ratio) for pair in conv_arch]
net = vgg(small_conv_arch)
lr, num_epochs, batch_size = 0.05, 10, 32
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)
d2l.train_ch6(net, train_iter, test_iter, num_epochs, lr, d2l.try_gpu())
training on cuda:0
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
<Figure size 350x250 with 1 Axes>
loss 0.219, train acc 0.921, test acc 0.912
392.3 examples/sec on cuda:0