第J2周：ResNet50V2 算法实战与解析

于 2024-12-26 20:48:54 发布
阅读量505 收藏 9
点赞数 9
CC 4.0 BY-SA版权
分类专栏： 机器学习365学习 文章标签： 人工智能 机器学习 python pytorch cnn 算法
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.youkuaiyun.com/DLearner/article/details/144752001

目录

FROM

我的环境

  • 语言环境：Python 3.11.9
  • 开发工具：Jupyter Lab
  • 深度学习环境：
    • torch==2.3.1+cu121
    • torchvision==0.18.1+cu121

一、本周内容和个人收获

1. v1和v2的区别

ResNet50V2是何恺明等人在ResNet基础上提出的改进版本，主要改进点在于残差结构和激活函数的使用。ResNetV2结构与原始ResNet结构的主要差别在于，原始ResNet结构先进行卷积操作，然后进行BN和激活函数计算，最后执行加法操作后再进行ReLU计算；而ResNetV2结构则是先进行BN和激活函数计算后卷积，把加法操作后的ReLU计算放到了残差结构内部。

ResNet v1 和 v2 的核心区别在于残差块（Residual Block）的内部结构：
在这里插入图片描述
1. 预激活设计：
v1：采用 “卷积层 -> BN层 -> ReLU” 的顺序
v2：采用 “BN层 -> ReLU -> 卷积层” 的顺序（预激活设计）
2. 信号流路径：
v1：identity分支在最后与主分支相加后再进行ReLU
v2：identity分支直接与主分支相加，不进行最后的ReLU

2. v2的主要结构

class ResNet50V2(nn.Module):
    def __init__(self, num_classes=1000):
        super(ResNet50V2, self).__init__()
        
        # 1. 初始层
        self.conv1 = nn.Sequential(
            nn.ZeroPad2d(3),
            nn.Conv2d(3, 64, kernel_size=7, stride=2),
            nn.ZeroPad2d(1),
            nn.MaxPool2d(kernel_size=3, stride=2)
        )
        
        # 2. 残差块堆叠
        self.conv2_x = self._make_layer(64, 64, blocks=3)    # 输出 256 通道
        self.conv3_x = self._make_layer(256, 128, blocks=4)  # 输出 512 通道
        self.conv4_x = self._make_layer(512, 256, blocks=6)  # 输出 1024 通道
        self.conv5_x = self._make_layer(1024, 512, blocks=3) # 输出 2048 通道
        
        # 3. 输出层
        self.post_bn = nn.BatchNorm2d(2048)
        self.post_relu = nn.ReLU(inplace=True)
        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(2048, num_classes)

二、核心代码及运行截图

模型代码：

import torch
import torch.nn as nn

class ResNet50V2(nn.Module):
    def __init__(self, 
                 include_top=True,      # 是否包含顶部的全连接层
                 preact=True,           # 是否使用预激活
                 use_bias=True,         # 是否使用偏置
                 input_shape=(3, 224, 224),  # 输入形状 (C,H,W)
                 pooling=None,          # 池化方式
                 num_classes=1000,      # 分类数量
                 classifier_activation='softmax'): # 分类层激活函数
        super(ResNet50V2, self).__init__()
        
        self.include_top = include_top
        self.preact = preact
        self.pooling = pooling
        
        # 初始层
        self.conv1_pad = nn.ZeroPad2d(3)
        self.conv1 = nn.Conv2d(input_shape[0], 64, kernel_size=7, 
                              stride=2, bias=use_bias)
        
        if not preact:
            self.conv1_bn = nn.BatchNorm2d(64)
            self.conv1_relu = nn.ReLU(inplace=True)
            
        self.pool1_pad = nn.ZeroPad2d(1)
        self.pool1 = nn.MaxPool2d(kernel_size=3, stride=2)
        
        # 残差块堆叠
        self.conv2 = self._make_stack2(64, 64, 3)
        self.conv3 = self._make_stack2(256, 128, 4)
        self.conv4 = self._make_stack2(512, 256, 6)
        self.conv5 = self._make_stack2(1024, 512, 3, stride1=1)
        
        if preact:
            self.post_bn = nn.BatchNorm2d(2048)
            self.post_relu = nn.ReLU(inplace=True)
            
        if include_top:
            self.avg_pool = nn.AdaptiveAvgPool2d((1, 1))
            self.fc = nn.Linear(2048, num_classes)
            if classifier_activation == 'softmax':
                self.activation = nn.Softmax(dim=1)
                
    def _make_stack2(self, in_channels, filters, blocks, stride1=2):
        layers = []
        layers.append(ResBlock(in_channels, filters, stride=stride1))
        for _ in range(1, blocks):
            layers.append(ResBlock(filters * 4, filters))
        return nn.Sequential(*layers)
        
    def forward(self, x):
        # 初始层
        x = self.conv1_pad(x)
        x = self.conv1(x)
        
        if not self.preact:
            x = self.conv1_bn(x)
            x = self.conv1_relu(x)
            
        x = self.pool1_pad(x)
        x = self.pool1(x)
        
        # 残差块
        x = self.conv2(x)
        x = self.conv3(x)
        x = self.conv4(x)
        x = self.conv5(x)
        
        if self.preact:
            x = self.post_bn(x)
            x = self.post_relu(x)
            
        if self.include_top:
            x = self.avg_pool(x)
            x = torch.flatten(x, 1)
            x = self.fc(x)
            if hasattr(self, 'activation'):
                x = self.activation(x)
        else:
            if self.pooling == 'avg':
                x = nn.functional.adaptive_avg_pool2d(x, (1, 1))
                x = torch.flatten(x, 1)
            elif self.pooling == 'max':
                x = nn.functional.adaptive_max_pool2d(x, (1, 1))
                x = torch.flatten(x, 1)
                
        return x

class ResBlock(nn.Module):
    def __init__(self, in_channels, filters, stride=1):
        super(ResBlock, self).__init__()
        
        self.bn1 = nn.BatchNorm2d(in_channels)
        self.conv1 = nn.Conv2d(in_channels, filters, kernel_size=1, stride=stride)
        
        self.bn2 = nn.BatchNorm2d(filters)
        self.conv2 = nn.Conv2d(filters, filters, kernel_size=3, stride=1, padding=1)
        
        self.bn3 = nn.BatchNorm2d(filters)
        self.conv3 = nn.Conv2d(filters, filters * 4, kernel_size=1)
        
        self.relu = nn.ReLU(inplace=True)
        
        # 快捷连接
        if stride != 1 or in_channels != filters * 4:
            self.shortcut = nn.Sequential(
                nn.Conv2d(in_channels, filters * 4, kernel_size=1, stride=stride)
            )
        else:
            self.shortcut = nn.Identity()
            
    def forward(self, x):
        identity = self.shortcut(x)
        
        out = self.bn1(x)
        out = self.relu(out)
        out = self.conv1(out)
        
        out = self.bn2(out)
        out = self.relu(out)
        out = self.conv2(out)
        
        out = self.bn3(out)
        out = self.relu(out)
        out = self.conv3(out)
        
        out += identity
        return out

# 创建模型
model = ResNet50V2(
    include_top=True,
    preact=True,
    input_shape=(3, 224, 224),
    num_classes=1000
)

model

ResNet50V2(
  (conv1_pad): ZeroPad2d((3, 3, 3, 3))
  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2))
  (pool1_pad): ZeroPad2d((1, 1, 1, 1))
  (pool1): MaxPool2d(kernel_size=3, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv2): Sequential(
    (0): ResBlock(
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(2, 2))
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Sequential(
        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(2, 2))
      )
    )
    (1): ResBlock(
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
    (2): ResBlock(
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
  )
  (conv3): Sequential(
    (0): ResBlock(
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Sequential(
        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2))
      )
    )
    (1): ResBlock(
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
    (2): ResBlock(
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
    (3): ResBlock(
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
  )
  (conv4): Sequential(
    (0): ResBlock(
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2))
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Sequential(
        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2))
      )
    )
    (1): ResBlock(
      (bn1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
    (2): ResBlock(
      (bn1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
    (3): ResBlock(
      (bn1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
    (4): ResBlock(
      (bn1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
    (5): ResBlock(
      (bn1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
  )
  (conv5): Sequential(
    (0): ResBlock(
      (bn1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Sequential(
        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(1, 1))
      )
    )
    (1): ResBlock(
      (bn1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
    (2): ResBlock(
      (bn1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
      (relu): ReLU(inplace=True)
      (shortcut): Identity()
    )
  )
  (post_bn): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (post_relu): ReLU(inplace=True)
  (avg_pool): AdaptiveAvgPool2d(output_size=(1, 1))
  (fc): Linear(in_features=2048, out_features=1000, bias=True)
  (activation): Softmax(dim=1)
)

from torchsummary import summary
import torch
from torch import nn

# 将模型移至 GPU（如果可用）
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)

# 使用 torchsummary 查看模型结构
summary(model, (3, 224, 224))

输出：

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
         ZeroPad2d-1          [-1, 3, 230, 230]               0
            Conv2d-2         [-1, 64, 112, 112]           9,472
         ZeroPad2d-3         [-1, 64, 114, 114]               0
         MaxPool2d-4           [-1, 64, 56, 56]               0
            Conv2d-5          [-1, 256, 28, 28]          16,640
       BatchNorm2d-6           [-1, 64, 56, 56]             128
              ReLU-7           [-1, 64, 56, 56]               0
            Conv2d-8           [-1, 64, 28, 28]           4,160
       BatchNorm2d-9           [-1, 64, 28, 28]             128
             ReLU-10           [-1, 64, 28, 28]               0
           Conv2d-11           [-1, 64, 28, 28]          36,928
      BatchNorm2d-12           [-1, 64, 28, 28]             128
             ReLU-13           [-1, 64, 28, 28]               0
           Conv2d-14          [-1, 256, 28, 28]          16,640
         ResBlock-15          [-1, 256, 28, 28]               0
         Identity-16          [-1, 256, 28, 28]               0
      BatchNorm2d-17          [-1, 256, 28, 28]             512
             ReLU-18          [-1, 256, 28, 28]               0
           Conv2d-19           [-1, 64, 28, 28]          16,448
      BatchNorm2d-20           [-1, 64, 28, 28]             128
             ReLU-21           [-1, 64, 28, 28]               0
           Conv2d-22           [-1, 64, 28, 28]          36,928
      BatchNorm2d-23           [-1, 64, 28, 28]             128
             ReLU-24           [-1, 64, 28, 28]               0
           Conv2d-25          [-1, 256, 28, 28]          16,640
         ResBlock-26          [-1, 256, 28, 28]               0
         Identity-27          [-1, 256, 28, 28]               0
      BatchNorm2d-28          [-1, 256, 28, 28]             512
             ReLU-29          [-1, 256, 28, 28]               0
           Conv2d-30           [-1, 64, 28, 28]          16,448
      BatchNorm2d-31           [-1, 64, 28, 28]             128
             ReLU-32           [-1, 64, 28, 28]               0
           Conv2d-33           [-1, 64, 28, 28]          36,928
      BatchNorm2d-34           [-1, 64, 28, 28]             128
             ReLU-35           [-1, 64, 28, 28]               0
           Conv2d-36          [-1, 256, 28, 28]          16,640
         ResBlock-37          [-1, 256, 28, 28]               0
           Conv2d-38          [-1, 512, 14, 14]         131,584
      BatchNorm2d-39          [-1, 256, 28, 28]             512
             ReLU-40          [-1, 256, 28, 28]               0
           Conv2d-41          [-1, 128, 14, 14]          32,896
      BatchNorm2d-42          [-1, 128, 14, 14]             256
             ReLU-43          [-1, 128, 14, 14]               0
           Conv2d-44          [-1, 128, 14, 14]         147,584
      BatchNorm2d-45          [-1, 128, 14, 14]             256
             ReLU-46          [-1, 128, 14, 14]               0
           Conv2d-47          [-1, 512, 14, 14]          66,048
         ResBlock-48          [-1, 512, 14, 14]               0
         Identity-49          [-1, 512, 14, 14]               0
      BatchNorm2d-50          [-1, 512, 14, 14]           1,024
             ReLU-51          [-1, 512, 14, 14]               0
           Conv2d-52          [-1, 128, 14, 14]          65,664
      BatchNorm2d-53          [-1, 128, 14, 14]             256
             ReLU-54          [-1, 128, 14, 14]               0
           Conv2d-55          [-1, 128, 14, 14]         147,584
      BatchNorm2d-56          [-1, 128, 14, 14]             256
             ReLU-57          [-1, 128, 14, 14]               0
           Conv2d-58          [-1, 512, 14, 14]          66,048
         ResBlock-59          [-1, 512, 14, 14]               0
         Identity-60          [-1, 512, 14, 14]               0
      BatchNorm2d-61          [-1, 512, 14, 14]           1,024
             ReLU-62          [-1, 512, 14, 14]               0
           Conv2d-63          [-1, 128, 14, 14]          65,664
      BatchNorm2d-64          [-1, 128, 14, 14]             256
             ReLU-65          [-1, 128, 14, 14]               0
           Conv2d-66          [-1, 128, 14, 14]         147,584
      BatchNorm2d-67          [-1, 128, 14, 14]             256
             ReLU-68          [-1, 128, 14, 14]               0
           Conv2d-69          [-1, 512, 14, 14]          66,048
         ResBlock-70          [-1, 512, 14, 14]               0
         Identity-71          [-1, 512, 14, 14]               0
      BatchNorm2d-72          [-1, 512, 14, 14]           1,024
             ReLU-73          [-1, 512, 14, 14]               0
           Conv2d-74          [-1, 128, 14, 14]          65,664
      BatchNorm2d-75          [-1, 128, 14, 14]             256
             ReLU-76          [-1, 128, 14, 14]               0
           Conv2d-77          [-1, 128, 14, 14]         147,584
      BatchNorm2d-78          [-1, 128, 14, 14]             256
             ReLU-79          [-1, 128, 14, 14]               0
           Conv2d-80          [-1, 512, 14, 14]          66,048
         ResBlock-81          [-1, 512, 14, 14]               0
           Conv2d-82           [-1, 1024, 7, 7]         525,312
      BatchNorm2d-83          [-1, 512, 14, 14]           1,024
             ReLU-84          [-1, 512, 14, 14]               0
           Conv2d-85            [-1, 256, 7, 7]         131,328
      BatchNorm2d-86            [-1, 256, 7, 7]             512
             ReLU-87            [-1, 256, 7, 7]               0
           Conv2d-88            [-1, 256, 7, 7]         590,080
      BatchNorm2d-89            [-1, 256, 7, 7]             512
             ReLU-90            [-1, 256, 7, 7]               0
           Conv2d-91           [-1, 1024, 7, 7]         263,168
         ResBlock-92           [-1, 1024, 7, 7]               0
         Identity-93           [-1, 1024, 7, 7]               0
      BatchNorm2d-94           [-1, 1024, 7, 7]           2,048
             ReLU-95           [-1, 1024, 7, 7]               0
           Conv2d-96            [-1, 256, 7, 7]         262,400
      BatchNorm2d-97            [-1, 256, 7, 7]             512
             ReLU-98            [-1, 256, 7, 7]               0
           Conv2d-99            [-1, 256, 7, 7]         590,080
     BatchNorm2d-100            [-1, 256, 7, 7]             512
            ReLU-101            [-1, 256, 7, 7]               0
          Conv2d-102           [-1, 1024, 7, 7]         263,168
        ResBlock-103           [-1, 1024, 7, 7]               0
        Identity-104           [-1, 1024, 7, 7]               0
     BatchNorm2d-105           [-1, 1024, 7, 7]           2,048
            ReLU-106           [-1, 1024, 7, 7]               0
          Conv2d-107            [-1, 256, 7, 7]         262,400
     BatchNorm2d-108            [-1, 256, 7, 7]             512
            ReLU-109            [-1, 256, 7, 7]               0
          Conv2d-110            [-1, 256, 7, 7]         590,080
     BatchNorm2d-111            [-1, 256, 7, 7]             512
            ReLU-112            [-1, 256, 7, 7]               0
          Conv2d-113           [-1, 1024, 7, 7]         263,168
        ResBlock-114           [-1, 1024, 7, 7]               0
        Identity-115           [-1, 1024, 7, 7]               0
     BatchNorm2d-116           [-1, 1024, 7, 7]           2,048
            ReLU-117           [-1, 1024, 7, 7]               0
          Conv2d-118            [-1, 256, 7, 7]         262,400
     BatchNorm2d-119            [-1, 256, 7, 7]             512
            ReLU-120            [-1, 256, 7, 7]               0
          Conv2d-121            [-1, 256, 7, 7]         590,080
     BatchNorm2d-122            [-1, 256, 7, 7]             512
            ReLU-123            [-1, 256, 7, 7]               0
          Conv2d-124           [-1, 1024, 7, 7]         263,168
        ResBlock-125           [-1, 1024, 7, 7]               0
        Identity-126           [-1, 1024, 7, 7]               0
     BatchNorm2d-127           [-1, 1024, 7, 7]           2,048
            ReLU-128           [-1, 1024, 7, 7]               0
          Conv2d-129            [-1, 256, 7, 7]         262,400
     BatchNorm2d-130            [-1, 256, 7, 7]             512
            ReLU-131            [-1, 256, 7, 7]               0
          Conv2d-132            [-1, 256, 7, 7]         590,080
     BatchNorm2d-133            [-1, 256, 7, 7]             512
            ReLU-134            [-1, 256, 7, 7]               0
          Conv2d-135           [-1, 1024, 7, 7]         263,168
        ResBlock-136           [-1, 1024, 7, 7]               0
        Identity-137           [-1, 1024, 7, 7]               0
     BatchNorm2d-138           [-1, 1024, 7, 7]           2,048
            ReLU-139           [-1, 1024, 7, 7]               0
          Conv2d-140            [-1, 256, 7, 7]         262,400
     BatchNorm2d-141            [-1, 256, 7, 7]             512
            ReLU-142            [-1, 256, 7, 7]               0
          Conv2d-143            [-1, 256, 7, 7]         590,080
     BatchNorm2d-144            [-1, 256, 7, 7]             512
            ReLU-145            [-1, 256, 7, 7]               0
          Conv2d-146           [-1, 1024, 7, 7]         263,168
        ResBlock-147           [-1, 1024, 7, 7]               0
          Conv2d-148           [-1, 2048, 7, 7]       2,099,200
     BatchNorm2d-149           [-1, 1024, 7, 7]           2,048
            ReLU-150           [-1, 1024, 7, 7]               0
          Conv2d-151            [-1, 512, 7, 7]         524,800
     BatchNorm2d-152            [-1, 512, 7, 7]           1,024
            ReLU-153            [-1, 512, 7, 7]               0
          Conv2d-154            [-1, 512, 7, 7]       2,359,808
     BatchNorm2d-155            [-1, 512, 7, 7]           1,024
            ReLU-156            [-1, 512, 7, 7]               0
          Conv2d-157           [-1, 2048, 7, 7]       1,050,624
        ResBlock-158           [-1, 2048, 7, 7]               0
        Identity-159           [-1, 2048, 7, 7]               0
     BatchNorm2d-160           [-1, 2048, 7, 7]           4,096
            ReLU-161           [-1, 2048, 7, 7]               0
          Conv2d-162            [-1, 512, 7, 7]       1,049,088
     BatchNorm2d-163            [-1, 512, 7, 7]           1,024
            ReLU-164            [-1, 512, 7, 7]               0
          Conv2d-165            [-1, 512, 7, 7]       2,359,808
     BatchNorm2d-166            [-1, 512, 7, 7]           1,024
            ReLU-167            [-1, 512, 7, 7]               0
          Conv2d-168           [-1, 2048, 7, 7]       1,050,624
        ResBlock-169           [-1, 2048, 7, 7]               0
        Identity-170           [-1, 2048, 7, 7]               0
     BatchNorm2d-171           [-1, 2048, 7, 7]           4,096
            ReLU-172           [-1, 2048, 7, 7]               0
          Conv2d-173            [-1, 512, 7, 7]       1,049,088
     BatchNorm2d-174            [-1, 512, 7, 7]           1,024
            ReLU-175            [-1, 512, 7, 7]               0
          Conv2d-176            [-1, 512, 7, 7]       2,359,808
     BatchNorm2d-177            [-1, 512, 7, 7]           1,024
            ReLU-178            [-1, 512, 7, 7]               0
          Conv2d-179           [-1, 2048, 7, 7]       1,050,624
        ResBlock-180           [-1, 2048, 7, 7]               0
     BatchNorm2d-181           [-1, 2048, 7, 7]           4,096
            ReLU-182           [-1, 2048, 7, 7]               0
AdaptiveAvgPool2d-183           [-1, 2048, 1, 1]               0
          Linear-184                 [-1, 1000]       2,049,000
         Softmax-185                 [-1, 1000]               0
================================================================
Total params: 25,575,912
Trainable params: 25,575,912
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 97.93
Params size (MB): 97.56
Estimated Total Size (MB): 196.07
----------------------------------------------------------------

训练结果：
在这里插入图片描述
在这里插入图片描述

ResNetV2模型复现--CVPR2016
冲出仁川，走出马德里，故事还会再继续
08-31 750
  图中的 Iterations 表示迭代次数；Test Error 表示测试集错误率。  (a)original 表示原始的 ResNet 的残差结构，(b)proposed 表示新的 ResNet 的残差结构。主要差别就是(a)结构先卷积后进行 BN 和激活函数计算，最后执行 addition 后再进行ReLU 计算； (b)结构先进行 BN 和激活函数计算后卷积，把 addition 后的 ReLU 计算放到了残差结构内部。  作者使用这两种不同的结构在 CIFAR-10 数据集上做测试，模型用的是
第12周：ResNet50V2算法实战与解析（Pytorch实现）
weixin_46620278的博客
05-29 1147
🍨 本文为[🔗365天深度学习训练营](https://mp.weixin.qq.com/s/0dvHCaOoFnW8SCp3JpzKxg) 中的学习记录博客🍖 原作者：[K同学啊](https://mtyjkh.blog.csdn.net/)说在前面本周学习目标：根据Tensflow的代码修改为对应的Pytorch代码；了解ResNetV2和ResNetV的区别我的环境：Python3.8、Pycharm2020、torch1.12.1+cu113。
resnet50v2
tjl521314_21的博客
03-20 1017
前面对resnet有了一定了解，主要关注其残差结构，其缓解了梯度消失的问题，是卷积神经网络的一大创举，这周主要学习resnet50v2,其是对resnet的改进。ResNet50V2 是对 ResNet50 的改进版本，它在保留了 ResNet 架构的基本特征的同时，进行了一些改动和优化，以提升模型性能和训练效率。
ResNet50V2
weixin_53657683的博客
07-03 1204
ResNetV2和ResNetV1都是深度残差网络（ResNet）的变体，它们的主要区别在于残差块的设计和批归一化（Batch Normalization, BN）的使用方式。ResNetV2是在ResNetV1的基础上进行改进的一种版本，旨在提高模型的性能和稳定性。在ResNetV1中，批归一化层位于每个卷积层的后面，ReLU 激活函数在批归一化之后。在ResNetV2中，批归一化和 ReLU 激活函数在每个卷积层之前进行。这种改变使得信息在模型中传播得更加顺畅，减轻了梯度消失的问题。
第J2周：ResNet50V2算法实战与解析
Pinospot的博客
02-17 579
ResNet50V2算法实战与解析
第J2周：ResNet50V2算法实战与解析（TensorFlow版）
weixin_43414521的博客
09-02 1276
- **🍨 本文为[]中的学习记录博客**>- **🍖 原作者：[]**– 1.请根据本文TensorFlow代码，编写出相应的Pytorch代码（建议使用上周的数据测试一下模型是否构建正确）– 2.了解ResNetV2与ResNetV的区别。
经典CNN算法解析实战-第J2周：ResNet50V2算法实战与解析
Oaix的博客
02-16 744
第J2周：ResNet50V2算法实战与解析
深度学习camp-第J2周：ResNet50V2算法实战与解析
最新发布
iracole的博客
11-25 536
ResNet50V2算法的构建
第J2周：ResNet50v2算法实战与解析
weixin_45748679的博客
04-26 734
size = len(dataloader.dataset) # 测试集的大小num_batches = len(dataloader) # 批次数目, (size/batch_size，向上取整)# 当不进行训练时，停止梯度更新，节省计算内存消耗# 计算loss本周手敲了ResNetV2算法，之前的模型框架没有发生改变，只是将模型方法进行了一个优化修改。残差块结构的简化： ResNetV2对残差块（Residual Block）的设计进行了简化。
resnet50-v2-7.onnx
10-24
resnet50 的onnx文件
resnet50二分类
11-01
resnet50二分类/recall_precision评价 包括数据读取，处理等函数
残差网络resnet_50_ckpt文件
06-10
resnet50已训练好模型参数，可在自己的数据上进行微调。tensorflow
ResNet50V2 网络结构
NLP与推荐算法
11-15 3102
来源于官方函数，input_shape=[160,160,3],classes=16 Layer (type) Output Shape Param # Connected to ========================================================...
J2：ResNet50v2算法实战与解析
zoey123456789的博客
11-01 1019
改进点：(a)original 表示原始的 ResNet 的残差结构，(b)proposed 表示新的 ResNet 的残差结构。从图中结果我们可以看出，(b)proposed 的测试集错误率明显更低一些，达到了 4.92%的错误率，(a)original 的测试集错误率是 7.61%。用不同 shortcut 结构的 ResNet-110 在CIFAR-10 数据集上做测试，发现最原始的(a)original 结构是最好的，也就是identity mapping 恒等映射是最好的。
深度学习-第J2周：ResNet-50-V2算法
weixin_43397208的博客
07-01 893
上周复现了何恺明在2015年提出的深度残差网络ResNet（deep residual network），大神在后来的论文中提到一种全新的残差单元，我们命名为ResNet_V2, 本文将在上周的基础上基于Pytorch实现ResNet_V2模型深度学习-第J1周：ResNet-50算法的Pytorch实现及解析_quant_day的博客-优快云博客。
深度学习J2周 ResNet50V2算法
Radiantsss的博客
11-21 739
了解ResNetV2的结构，并对比与ResNet的不同将tensflow代码转换为pytorch具体的细节还得再去读一下论文， 有些专有名词还是不太懂。
基于tensorflow的ResNet50V2网络识别动物
11-29 2404
之前很多人在，如何进行XXX的识别，对应的神经网络如何搭建。对应神经网络怎么搭建，我也是照本宣科，只能说看得懂而已，没有对这块进行深入的研究，但是现在tensorflow，paddle这些工具，都提供了非常成熟的神经网络进行直接使用。本文对过往的一些文章进行改造，使用已经集成的神经网络，简单的实现多个种类的动物识别。
深度学习第J2周：ResNet50V2算法实战与解析
m0_62237233的博客
03-19 2839
何恺明大神在这篇论文中提出了一种新的残差单元。我们将这篇论文中的ResNet结构称为ResNetV2。
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值