AlexNet原理及代码实现

最新推荐文章于 2024-09-18 23:38:35 发布

殇小气

最新推荐文章于 2024-09-18 23:38:35 发布

阅读量506

点赞数

分类专栏： python深度学习记录文章标签：卷积卷积神经网络

本文链接：https://blog.csdn.net/qq_51778415/article/details/115272685

版权

python深度学习记录专栏收录该内容

13 篇文章 0 订阅

订阅专栏

AlexNet 原理及代码实现

基础介绍：

原始图像：256X256X3
图像处理：数据增强
1. 随机剪切256X256x3<=>224x224x3
2. 224x224旋转，位置变换
3. 图像增大：224x224<=>227x227
4. 实际输入AlexNet网络：227x227x3（可以通过padding实现）
输入图像为3通道，之后采用双GPU并行运算

基础步骤：

卷积层---->relu---->池化层---->norm层---->全连接层

总分为5个卷积，3个全连接
输入为：227x227x3
conv1:
1. filter:11x11x3
2. padding: 0
3. stride: 4
4. depth: 48x2 = 96

max pooling:
1. filter:3x3
2. stride:2

conv2:
1. filter:5x5x48
2. padding: 2
3. stride:1
4. depth:128x2 =256

max pooling:
1. filter:3x3
2. stride:2

conv3:
1. filter:3x3x256
2. padding: 1
3. stride:1
4. depth:192x2 = 384

conv4:
1. filter:3x3x192
2. padding:1
3. stride:1
4. depth:192x2 = 384

conv5:
1. filter:3x3x192
2. padding:1
3. stride:1
4. depth:128x2

max pooling:
1. filter:3x3
2. stride:2

fc1:
6x6x128x2=9216==>2048x2
fc2:
2048x2=4096==>2048x2
fc2:
4096==>1000
ReLU：线性归一
norm层：去均值，幅度归一化

import torch
import torch.nn as nn
class AlexNet(nn.Module):
    def __init__(self, num_classes=1000):
        self.featrues = nn.Sequential(
            # conv1
            nn.Conv2d(3, 96, kernel_size=11, stride=4),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2),
            nn.BatchNorm2d(96),
            # conv2
            nn.Conv2d(96, 256, kernel_size=5, padding=2),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2),
            nn.BatchNorm2d(256),
            # conv3
            nn.Conv2d(256, 384, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            # conv4
            nn.Conv2d(384, 384, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            # conv5
            nn.Conv2d(384, 256, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2))

        self.classifier = nn.Sequential(
            # fc1
            nn.Linear(256 * 6 * 6, 4096),
            nn.ReLU(inplace=True),
            # fc2
            nn.Linear(4096, 4096),
            nn.ReLU(inplace=True),
            # fc3
            nn.Linear(4096, num_classes))

        def forward(self, x):
        	#参数过滤调整
            x = self.features(x)
            x = x.view(x.size(0), 256 * 6 * 6)
            x = self.classifier(x)
            return x

net=AlexNet()