第四章分类任务

whilecontinue

已于 2025-01-13 23:16:02 修改

阅读量1.9k

点赞数 49

分类专栏：李哥深度学习项目班文章标签：分类人工智能深度学习

于 2025-01-13 23:15:09 首次发布

本文链接：https://blog.youkuaiyun.com/whilecontinue/article/details/145099567

版权

所有内容均来源于【李哥考研复试项目班】

一、AlexNet

1、特征图维度变化

输入图片 3 * 244 * 244

输出参数量

Conv2d(3, 64, 11, 4, 2) (224-11+2*2)/4+1=55 64*3*11*11+64(偏置) =

64 * 55 * 55 23296

Pool(3, 2) 64 * 27 * 27 0

3个一pool，每次走两步

Conv2d(64, 192, 5, 1, 2) (27-5+2*2)/1+1=27 192*64*5*5+64=

192 * 27 * 27 307392

Pool(3, 2) 192 * 13 * 13 0

Conv2d(192, 384, 3, 1, 1) (13-3+2*1)/1+1=13 384*192*3*3+384=

384 * 13 * 13 663936

Conv2d(384, 256, 3, 1, 1) (13-3+2*1)/1+1=13 256*384*3*3+256=

256 * 13 * 13 884992

Conv2d(256, 256, 3, 1, 1) (13-3+2*1)/1+1=13 256*256*3*3*+256=

256 * 13 * 13 590080

Pool(3, 2) 256 * 6 * 6 0

adaPool(6) 256 * 6 * 6 0

拉直 256*6*6=9216个数

Linner(9216, 4096) 4096 9216*4096+4096=37752832

Linner(4096, 4096) 4096 4096*4096+4096=16781312

Linner(4096, 1000) 1000 4096*1000+1000=4097000

2、代码

import torch
import torchvision.models as models
import torch.nn as nn


class myAlexNet(nn.Module):
    def __init__(self, out_dim):
        super(myAlexNet, self).__init__()
        self.conv1 = nn.Conv2d(3, 64, 11, 4, 2)
        # 定义了一个卷积层，3个输入通道，64个输出通道，卷积核大小为11x11，步长为4，填充为2
        self.pool1 = nn.MaxPool2d(3, 2)
        # 定义池化层，赤化窗口大小为3*3，步长为2

        self.conv2 = nn.Conv2d(64, 192, 5, 1, 2)
        self.pool2 = nn.MaxPool2d(3, 2)

        self.conv3 = nn.Conv2d(192, 384, 3, 1, 1)
        self.conv4 = nn.Conv2d(384, 256, 3, 1, 1)
        self.conv5 = nn.Conv2d(256, 256, 3, 1, 1)

        self.pool3 = nn.MaxPool2d(3, 2)
        self.pool4 = nn.AdaptiveAvgPool2d(6)

        self.fc1 = nn.Linear(9216, 4096)
        # 定义全连接层，将卷积层的