第12周：ResNet50V2算法实战与解析（Pytorch实现）

目录

前言

一、ResNetV2介绍

1.1 ResNetV2结构与ResNet结构对比

1.2 关于残差结构的不同尝试

1.3 关于激活的尝试

二、模型复现

2.1 数据导入

2.2 模型构建

2.2.1 Residual Block

2.2.2 堆叠Residual Block

2.2.3 ResNet50V2架构搭建

三、模型训练

四、结果可视化

五、指定图片预测

总结

---

前言

• 🍨 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

数据来源：[K同学啊](https://mtyjkh.blog.csdn.net/)

---

一、ResNetV2介绍

ResNetV2是何凯明在下面这篇论文中提出的一种新的残差单元

1.1 ResNetV2结构与ResNet结构对比

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；主要的差别就是原来的结构是先卷积后进行BN和激活函数计算，最后在执行addition后再进行ReLU计算，而新的结构是先进行了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。

改进结果：作者使用这两种不同的结构再CIFAR-10数据集上做测试，模型用的是1001层的ResNet模型，从上图中的右图我们可以看出，proposed模型的测试集错误率明显更低一些，达到1了4.92%的错误率，而原始的结构的测试集错误率是7.61%

1.2 关于残差结构的不同尝试

如上图所示，图(b-f)中的快捷连接被不同的组件阻碍，这里为简化插图，不显示BN层，所有单位均采用权值层之后的BN层。图中(a-f)都是作者对于残差结构的shortcut部分进行的不同尝试，它们对应的实验结果如下表所示：

从上表的结果中可以发现，最原始的(a)original结构是最好的，也就是idengtity mapping恒等映射是最好的

1.3 关于激活的尝试

可以看出，不同的激活方式，最好的结果是(e)full pre-activation，其次是(a)original

二、模型复现

2.1 数据导入

这里实验的数据集仍采用的是上周鸟类识别的数据，除了模型具体结构有变化，其余都和上周的代码一致

参看ResNet50这篇文章：第11周：ResNet-50算法实战与解析（Pytorch实现）-CSDN博客

代码如下：

import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets
import os,PIL,pathlib,warnings
import torch.nn.functional as F
import matplotlib.pyplot as plt
import pandas as pd
from torchvision.io import read_image
from torch.utils.data import Dataset
import torch.utils.data as data
from PIL import Image
import copy
import numpy as np


#一、导入数据
'''
1.1 设置GPU
'''
warnings.filterwarnings("ignore")

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)

'''
1.2 导入数据
'''
plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False  # 用来正常显示负号
data_dir = './bird_photos/'
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob('*'))
def count_images(folder):
    count = 0
    for item in folder.iterdir():
        if item.is_file():
            count += 1
        if item.is_dir():
            count += count_images(item)
    return count

image_count = count_images(data_dir)
print("图片总数为：", image_count)
classNames = [str(path).split('\\')[1] for path in data_paths]
#利用split()函数对data_paths中的每个文件路径执行分割操作，获取各个文件所属的类别名称并储存在classNames中
# 4类天气，各300张图片
print(classNames)

#二、数据预处理
'''
2.1 加载数据
'''
train_transforms = transforms.Compose([
    transforms.Resize([224, 224]),  # 将输入图片resize成统一尺寸
    # transforms.RandomHorizontalFlip(), # 随机水平翻转
    transforms.ToTensor(),          # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
    transforms.Normalize(           # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225])  # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])

test_transform = transforms.Compose([
    transforms.Resize([224, 224]),  # 将输入图片resize成统一尺寸
    transforms.ToTensor(),          # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
    transforms.Normalize(           # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225])  # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])

total_data = datasets.ImageFolder("./bird_photos/",transform=train_transforms)
print(total_data)
print(total_data.class_to_idx)

'''
2.2 划分数据集
'''
train_size = int(0.8 * len(total_data))
test_size = len(total_data) - train_size
train_dataset, test_dataset = torch.utils.data.random_split(total_data, [train_size, test_size])
print(train_dataset, test_dataset)

'''
2.3 可视化数据
'''
batch_size = 8
train_dl = torch.utils.data.DataLoader(train_dataset,
                                       batch_size=batch_size,
                                       shuffle=True,
                                       num_workers=0)
test_dl = torch.utils.data.DataLoader(test_dataset,
                                      batch_size=batch_size,
                                      shuffle=True,
                                      num_workers=0)
for X, y in test_dl:
    print("Shape of X [N, C, H, W]: ", X.shape)
    print("Shape of y: ", y.shape, y.dtype)
    break

image_folder = './bird_photos/Cockatoo/'           #指定图像文件夹路径

image_files = [f for f in os.listdir(image_folder) if f.endswith((".jpg", ".png", ".jpeg"))]
fig, axes = plt.subplots(2, 4, figsize=(16, 6))

for ax, img_file in zip(axes.flat, image_files):
    img_path = os.path.join(image_folder, img_file)
    img = Image.open(img_path)
    ax.imshow(img)
    ax.axis('off')
plt.tight_layout()
plt.show()

2.2 模型构建

ResNet50V2、ResNet101V2与ResNet152V2的搭建方式完全一样，区别就在于堆叠的Residual Block的数量不同

2.2.1 Residual Block

与ResNet50在identity shortcut这里是有区别的，在ResNet50中是这样的

if conv_shortcut:

    self.shortcut = nn.Sequential( nn.Conv2d(in_channels, 4 * filters, kernel_size=1,       stride=stride, bias=False), nn.BatchNorm2d(4 * filters, eps=1.001e-5) )

代码如下：

'''
3.1 residual block
'''
class Block2(nn.Module):
    def __init__(self, in_channels, filters, kernel_size=3, stride=1, conv_shortcut=False):
        super(Block2, self).__init__()
        self.conv_shortcut = conv_shortcut
        self.stride = stride

        self.bn_preact = nn.BatchNorm2d(in_channels)
        self.relu_preact = nn.ReLU(inplace=True)

        if conv_shortcut:
            self.shortcut = nn.Conv2d(in_channels, 4 * filters, kernel_size=1, stride=stride)
        else:
            self.shortcut = nn.MaxPool2d(kernel_size=1, stride=stride) if stride > 1 else nn.Identity()

        self.conv1 = nn.Conv2d(in_channels, filters, kernel_size=1, stride=1, bias=False)
        self.bn1 = nn.BatchNorm2d(filters)
        self.relu1 = nn.ReLU(inplace=True)

        self.pad2 = nn.ZeroPad2d(1)
        self.conv2 = nn.Conv2d(filters, filters, kernel_size=kernel_size, stride=stride, bias=False)
        self.bn2 = nn.BatchNorm2d(filters)
        self.relu2 = nn.ReLU(inplace=True)

        self.conv3 = nn.Conv2d(filters, 4 * filters, kernel_size=1)

    def forward(self, x):
        preact = self.bn_preact(x)
        preact = self.relu_preact(preact)

        shortcut = self.shortcut(preact)

        x = self.conv1(preact)
        x = self.bn1(x)
        x = self.relu1(x)

        x = self.pad2(x)
        x = self.conv2(x)
        x = self.bn2(x)
        x = self.relu2(x)

        x = self.conv3(x)

        x += shortcut
        return x

2.2.2 堆叠Residual Block

代码如下：

class Stack2(nn.Module):
    def __init__(self, in_channels, filters, blocks, stride=2):
        super(Stack2, self).__init__()
        self.blocks = nn.ModuleList()
        self.blocks.append(Block2(in_channels, filters, conv_shortcut=True))
        for i in range(1, blocks - 1):
            self.blocks.append(Block2(4 * filters, filters))
        self.blocks.append(Block2(4 * filters, filters, stride=stride))

    def forward(self, x):
        for block in self.blocks:
            x = block(x)
        return x

2.2.3 ResNet50V2架构搭建

网络结构示意图如下

搭建代码如下：

class ResNet50V2(nn.Module):
    def __init__(self,
                 include_top=True,  # 是否包含位于网络顶部的全链接层
                 preact=True,  # 是否使用预激活
                 use_bias=True,  # 是否对卷积层使用偏置
                 input_shape=[224, 224, 3],
                 classes=1000,
                 pooling=None):  # 用于分类图像的可选类数
        super(ResNet50V2, self).__init__()

        self.conv1 = nn.Sequential()
        self.conv1.add_module('conv', nn.Conv2d(3, 64, 7, stride=2, padding=3, bias=use_bias, padding_mode='zeros'))
        if not preact:
            self.conv1.add_module('bn', nn.BatchNorm2d(64))
            self.conv1.add_module('relu', nn.ReLU())
        self.conv1.add_module('max_pool', nn.MaxPool2d(kernel_size=3, stride=2, padding=1))

        self.conv2 = Stack2(64, 64, 3)
        self.conv3 = Stack2(256, 128, 4)
        self.conv4 = Stack2(512, 256, 6)
        self.conv5 = Stack2(1024, 512, 3, stride=1)

        self.post = nn.Sequential()
        if preact:
            self.post.add_module('bn', nn.BatchNorm2d(2048))
            self.post.add_module('relu', nn.ReLU())
        if include_top:
            self.post.add_module('avg_pool', nn.AdaptiveAvgPool2d((1, 1)))
            self.post.add_module('flatten', nn.Flatten())
            self.post.add_module('fc', nn.Linear(2048, classes))
        else:
            if pooling == 'avg':
                self.post.add_module('avg_pool', nn.AdaptiveAvgPool2d((1, 1)))
            elif pooling == 'max':
                self.post.add_module('max_pool', nn.AdaptiveMaxPool2d((1, 1)))

    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        x = self.conv4(x)
        x = self.conv5(x)
        x = self.post(x)
        return x

model = ResNet50V2().to(device)
print(model)

模型打印输出如下：

ResNet50V2(
  (conv1): Sequential(
    (conv): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3))
    (max_pool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
  )
  (conv2): Stack2(
    (blocks): ModuleList(
      (0): Block2(
        (bn_preact): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
        (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), bias=False)
        (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
      )
      (1): Block2(
        (bn_preact): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): Identity()
        (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), bias=False)
        (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
      )
      (2): Block2(
        (bn_preact): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): MaxPool2d(kernel_size=1, stride=2, padding=0, dilation=1, ceil_mode=False)
        (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(2, 2), bias=False)
        (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))
      )
    )
  )
  (conv3): Stack2(
    (blocks): ModuleList(
      (0): Block2(
        (bn_preact): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): Conv2d(256, 512, kernel_size=(1, 1), stride=(1, 1))
        (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), bias=False)
        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
      )
      (1-2): 2 x Block2(
        (bn_preact): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): Identity()
        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), bias=False)
        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
      )
      (3): Block2(
        (bn_preact): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): MaxPool2d(kernel_size=1, stride=2, padding=0, dilation=1, ceil_mode=False)
        (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), bias=False)
        (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))
      )
    )
  )
  (conv4): Stack2(
    (blocks): ModuleList(
      (0): Block2(
        (bn_preact): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): Conv2d(512, 1024, kernel_size=(1, 1), stride=(1, 1))
        (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), bias=False)
        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
      )
      (1-4): 4 x Block2(
        (bn_preact): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): Identity()
        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), bias=False)
        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
      )
      (5): Block2(
        (bn_preact): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): MaxPool2d(kernel_size=1, stride=2, padding=0, dilation=1, ceil_mode=False)
        (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), bias=False)
        (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))
      )
    )
  )
  (conv5): Stack2(
    (blocks): ModuleList(
      (0): Block2(
        (bn_preact): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(1, 1))
        (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), bias=False)
        (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
      )
      (1-2): 2 x Block2(
        (bn_preact): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu_preact): ReLU(inplace=True)
        (shortcut): Identity()
        (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
        (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu1): ReLU(inplace=True)
        (pad2): ZeroPad2d((1, 1, 1, 1))
        (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), bias=False)
        (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
        (relu2): ReLU(inplace=True)
        (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1))
      )
    )
  )
  (post): Sequential(
    (bn): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu): ReLU()
    (avg_pool): AdaptiveAvgPool2d(output_size=(1, 1))
    (flatten): Flatten(start_dim=1, end_dim=-1)
    (fc): Linear(in_features=2048, out_features=1000, bias=True)
  )
）

三、模型训练

训练函数和测试函数都与前面的文章中一致

代码如下：

'''
4.1 编写训练函数
'''
def train(dataloader, model, optimizer, loss_fn):
    size = len(dataloader.dataset)
    num_batches = len(dataloader)

    train_acc, train_loss = 0, 0

    for X, y in dataloader:
        X, y = X.to(device), y.to(device)

        pred = model(X)
        loss = loss_fn(pred, y)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        train_loss += loss.item()
        train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()

    train_loss /= num_batches
    train_acc /= size

    return train_acc, train_loss

'''
4.2 编写测试函数
'''
def test(dataloader, model, loss_fn):
    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)
    test_loss, test_acc = 0, 0

    # 当不进行训练时，停止梯度更新，节省计算内存消耗
    with torch.no_grad():
        for imgs, target in dataloader:
            imgs, target = imgs.to(device), target.to(device)

            # 计算loss
            target_pred = model(imgs)
            loss = loss_fn(target_pred, target)

            test_loss += loss.item()
            test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item()

    test_acc /= size
    test_loss /= num_batches

    return test_acc, test_loss

'''
4.3 正式训练
'''

loss_fn = nn.CrossEntropyLoss()   #交叉熵函数
learn_rate = 1e-3
opt = torch.optim.Adam(model.parameters(), lr=learn_rate)

epochs = 30
train_loss = []
train_acc = []
test_loss = []
test_acc = []
best_acc = 0

# 开始训练
for epoch in range(epochs):

    model.train()
    epoch_train_acc, epoch_train_loss = train(train_dl, model, opt, loss_fn)

    model.eval()
    epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)

    if epoch_test_acc > best_acc:
        best_acc = epoch_test_acc
        best_model = copy.deepcopy(model)

    train_acc.append(epoch_train_acc)
    train_loss.append(epoch_train_loss)
    test_acc.append(epoch_test_acc)
    test_loss.append(epoch_test_loss)

    lr = opt.state_dict()['param_groups'][0]['lr']

    template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}')
    print(template.format(epoch + 1, epoch_train_acc * 100, epoch_train_loss,
                          epoch_test_acc * 100, epoch_test_loss, lr))

print('Done')

训练过程如下：

四、结果可视化

代码如下：

#五、结果可视化
import warnings
warnings.filterwarnings("ignore")             #忽略警告信息
plt.rcParams['font.sans-serif'] = ['SimHei']  #用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False    #用来正常显示负号
plt.rcParams['figure.dpi'] = 100              #分辨率

epochs_range = range(epochs)

plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)

plt.plot(epochs_range, train_acc, label='Training Accuracy')
plt.plot(epochs_range, test_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')

plt.subplot(1, 2, 2)
plt.plot(epochs_range, train_loss, label='Training Loss')
plt.plot(epochs_range, test_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()

打印输出如下：

五、指定图片预测

#六、指定图片预测
classes = list(total_data.class_to_idx)
def predict_one_image(image_path, model, transform, classes):
    test_img = Image.open(image_path).convert('RGB')
    plt.imshow(test_img)  # 展示预测的图片

    test_img = transform(test_img)

    img = test_img.to(device).unsqueeze(0)

    model.eval()
    output = model(img)

    _, pred = torch.max(output, 1)
    pred_class = classes[pred]
    print(f'预测结果是：{pred_class}')

    # 预测训练集中的某张照片

predict_one_image(image_path='./bird_photos/Bananaquit/008.jpg',
                  model=model,
                  transform=train_transforms,
                  classes=classes)

输出结果：Bananaquit

预测正确

---

总结

• 了解了ResNetV2的结构，对比了与原来的改变之处
• 实践了将tensflow代码转换为pytorch
• 验证了所搭建模型的可行性，但同时发现整个训练过程并不是很稳定，测试集上最高准确率能达到91.2%