AlexNet——训练花数据集

目录

一、网络结构

二、创新点分析

三、知识点

1. nn.ReLU(inplace) 

2. os.getcwd与os.path.abspath 

3. 使用torchvision下的datasets包 

4. items()与dict()用法 

5. json文件  

6. tqdm

7. net.train()与net.val()

四、代码

---

AlexNet是由Alex Krizhevsky、Ilya Sutskever和Geoffrey Hinton在2012年ImageNet图像分类竞赛中提出的一种经典的卷积神经网络。AlexNet使用了Dropout层，减少过拟合现象的发生。

一、网络结构

二、数据集 

文件存放：

dataset->flower_data->flower_photos

再使用split_data.py 将数据集根据比例划分成训练集和预测集

详细请查看b站up主霹雳吧啦Wz：https://github.com/WZMIAOMIAO/deep-learning-for-image-processing/blob/master/pytorch_classification

三、创新点分析

1. deeper网络结构

    通过增加网络深度，AlexNet可以更好的学习数据集的特征，并提高分类的准确率。

2. 使用ReLU激活函数，克服梯度消失以及求梯度复杂的问题。

3. 使用LRN局部响应归一化

    LRN是在卷积与池化层间添加归一化操作。卷积过程中，每个卷积核都对应一个feature map，LRN对这些feature map进行归一化操作。即，对每个特征图的每个位置，计算该位置周围的像素平方和，然后将当前位置像素值除以这个和。LRN可抑制邻近神经元的响应，在一定程度上能够避免过拟合，提高网络泛化能力。

4. 使用Dropout层

Dropout层：在训练过程中随机删除一定比例的神经元，以减少过拟合。Dropout一般放在全连接层与全连接层之间。

四、知识点

1. nn.ReLU(inplace) 默认参数为：inplace=False

inplace=False：不会修改输入对象的值，而是返回一个新创建的对象，即打印出的对象存储地址不同。（值传递）

inplace=True：会修改输入对象的值，即打印的对象存储地址相同，可以节省申请与释放内存的空间与时间。（地址传递）

import torch
import numpy as np
import torch.nn as nn
 
# id()方法返回对象的内存地址
relu1 = nn.ReLU(inplace=False)
relu2 = nn.ReLU(inplace=True)
data = np.random.randn(2, 4)
input = torch.from_numpy(data)  # 转换成tensor类型
print("input address:", id(input))
output1 = relu1(input)
print("replace=False -- output address:", id(output1))
output2 = relu2(input)
print("replace=True -- output address:", id(output2))
# input address: 1669839583200
# replace=False -- output address: 1669817512352
# replace=True -- output address: 1669839583200

2. os.getcwd与os.path.abspath 

os.getcwd()：获取当前工作目录

os.path.abspath('xxx.py')：获取文件当前的完整路径

import os
 
print(os.getcwd())  # D:\Code
print(os.path.abspath('test.py'))  # D:\Code\test.py

3. 使用torchvision下的datasets包 

train_dataset=datasets.ImageFolder(root=os.path.join(image_path,'train'),transform=data_transform['train'])

可以得出这些信息： 

4. items()与dict()用法 

items()：把字典中的每对key和value组成一个元组，并将这些元组放在列表中返回。

obj = {
    'dog': 0,
    'cat': 1,
    'fish': 2
}
print(obj)  # {'dog': 0, 'cat': 1, 'fish': 2}
print(obj.items())  # dict_items([('dog', 0), ('cat', 1), ('fish', 2)])
print(dict((v, k) for k, v in obj.items()))  # {0: 'dog', 1: 'cat', 2: 'fish'}

5. json文件  

（1）json.dumps：将Python对象编码成JSON字符串

（2）json.loads：将已编码的JSON字符串编码为Python对象

import json
 
data = [1, 2, 3]
data_json = json.dumps(data)  # 
data = json.loads(data_json)
print(type(data))  # 

6. tqdm

train_bar = tqdm(train_loader, file=sys.stdout)
使用tqdm函数，对train_loader进行迭代，将进度条输出到标准输出流sys.stdout中。可以方便用户查看训练进度。

from tqdm import tqdm
import time
 
for i in tqdm(range(10)):
    time.sleep(0.1)

7. net.train()与net.val()

net.train()：启用BatchNormalization和Dropout

net.eval|()：不启用BatchNormalization和Dropout

五、代码

model.py

import torch
import torch.nn as nn
 
class AlexNet(nn.Module):
    def __init__(self, num_classes=1000):
        super(AlexNet, self).__init__()
        self.features = nn.Sequential(
            nn.Conv2d(3, 96, kernel_size=11, padding=2, stride=4),  # input[3,224,224] output[96,55,55]
            nn.ReLU(inplace=True),  # inplace=True 址传递
            nn.MaxPool2d(kernel_size=3, stride=2),  # output[96,27,27]
            nn.Conv2d(96, 256, kernel_size=5, padding=2),  # output[256,27,27]
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2),  # output[256,13,13]
            nn.Conv2d(256, 384, kernel_size=3, padding=1),  # output[384,13,13]
            nn.ReLU(inplace=True),
            nn.Conv2d(384, 384, kernel_size=3, padding=1),  # output[384,13,13]
            nn.ReLU(inplace=True),
            nn.Conv2d(384, 256, kernel_size=3, padding=1),  # output[256,13,13]
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2),  # output[256,6,6]
        )
        self.classifier = nn.Sequential(
            nn.Dropout(p=0.5),
            nn.Linear(256 * 6 * 6, 2048),
            nn.ReLU(inplace=True),
            nn.Dropout(p=0.5),
            nn.Linear(2048, 2048),
            nn.ReLU(inplace=True),
            nn.Linear(2048, num_classes)
        )
 
    def forward(self, x):
        x = self.features(x)
        x = torch.flatten(x, start_dim=1)  # batch这一维度不用，从channel开始
        x = self.classifier(x)
        return x

train.py 

import os
import torch
import torch.nn as nn
from torchvision import transforms, datasets
import json
from model import AlexNet
import torch.optim as optim
from tqdm import tqdm
 
def main():
    device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
    print("using:{}".format(device))
    data_transform = {
        'train': transforms.Compose([
            # 将给定图像随机裁剪为不同的大小和宽高比，然后缩放所裁剪得到的图像为指定大小
            transforms.RandomResizedCrop(224),
            # 水平方向随机翻转
            transforms.RandomHorizontalFlip(),
            transforms.ToTensor(),
            transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
        ]),
        'val': transforms.Compose([
            transforms.Resize((224, 224)),
            transforms.ToTensor(),
            transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
        ])
    }
    # get data root path
    data_root = os.path.abspath(os.getcwd())  # D:\Code\AlexNet
    # get flower data set path
    image_path = os.path.join(data_root, 'data_set', 'flower_data')  # D:\Code\AlexNet\data_set\flower_data
    # 使用assert断言语句：出现错误条件时，就触发异常
    assert os.path.exists(image_path), '{} path does not exist!'.format(image_path)
 
    train_dataset = datasets.ImageFolder(root=os.path.join(image_path, 'train'), transform=data_transform['train'])
    val_dataset = datasets.ImageFolder(root=os.path.join(image_path, 'val'), transform=data_transform['val'])
    train_num = len(train_dataset)
    val_num = len(val_dataset)
 
    # write class_dict into json file
    flower_list = train_dataset.class_to_idx
    class_dict = dict((v, k) for k, v in flower_list.items())
    json_str = json.dumps(class_dict)
    with open('class_indices.json', 'w') as file:
        file.write(json_str)
 
    batch_size = 32
    train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=0)
    val_loader = torch.utils.data.DataLoader(val_dataset, batch_size=4, shuffle=False, num_workers=0)
 
    net = AlexNet(num_classes=5)
    net.to(device)
    loss_function = nn.CrossEntropyLoss()
    optimizer = optim.Adam(net.parameters(), lr=0.0002)
    epochs = 5
    save_path = './model/AlexNet.pth'
    best_acc = 0.0
    train_steps = len(train_loader)  # train_num / batch_size
    train_bar = tqdm(train_loader)
    val_bar = tqdm(val_loader)
 
    for epoch in range(epochs):
        # train
        net.train()
        epoch_loss = 0.0
        # 加入进度条
        train_bar = tqdm(train_loader)
        for step, data in enumerate(train_bar):
            images, labels = data
            optimizer.zero_grad()
            outputs = net(images.to(device))
            loss = loss_function(outputs, labels.to(device))
            loss.backward()
            optimizer.step()  # update x by optimizer
 
            # print statistics
            epoch_loss += loss.item()
            train_bar.desc = 'train eporch[{}/{}] loss:{:.3f}'.format(epoch + 1, epochs, loss)
 
        # validate
        net.eval()
        acc = 0.0
        with torch.no_grad():
            val_bar = tqdm(val_loader)
            for val_data in val_bar:
                val_images, val_labels = val_data
                outputs = net(val_images.to(device))
                predict_y = torch.max(outputs, dim=1)[1]  # [1]取每行最大值的索引
                acc += torch.eq(predict_y, val_labels.to(device)).sum().item()
        val_acc = acc / val_num
        print('[epoch %d] train_loss: %.3f val_accuracy: %.3f' % (epoch + 1, epoch_loss / train_steps, val_acc))
 
        # find best accuracy
        if val_acc > best_acc:
            best_acc = val_acc
            torch.save(net.state_dict(), save_path)
    print('Train finished!')
 
 
if __name__ == '__main__':
    main()

class_indices.json

{"0": "daisy", "1": "dandelion", "2": "roses", "3": "sunflowers", "4": "tulips"}

predict.py 

import os
import torch
from torchvision import transforms
from PIL import Image
import matplotlib.pyplot as plt
import json
from model import AlexNet
 
def main():
    device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
    transform = transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor(),
        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
    ])
 
    # load image
    img_path = './2.jpg'
    assert os.path.exists(img_path), "file:'{}' does not exist".format(img_path)
    img = Image.open(img_path)
    plt.imshow(img)
 
    # input [N,C,H,W]
    img = transform(img)
    img = torch.unsqueeze(img, dim=0)
 
    # read class_indices
    json_path = './class_indices.json'
    assert os.path.exists(json_path), "file:'{}' does not exist".format(json_path)
    with open(json_path, 'r') as file:
        class_dict = json.load(file)  # {'0': 'daisy', '1': 'dandelion', '2': 'roses', '3': 'sunflowers', '4': 'tulips'}
 
    # load model
    net = AlexNet(num_classes=5).to(device)
    # load model weights
    weight_path = './model/AlexNet.pth'
    assert os.path.exists(weight_path), "file:'{}' does not exist".format(weight_path)
    net.load_state_dict(torch.load(weight_path))
 
    # predict
    net.eval()
    with torch.no_grad():
        output = torch.squeeze(net(img.to(device))).cpu()
        predict = torch.softmax(output, dim=0)
        predict_class = torch.argmax(predict).numpy()
    print_res = 'class:{} probability:{:.3}'.format(class_dict[str(predict_class)], predict[predict_class].numpy())
    plt.title(print_res)
    plt.show()
 
    for i in range(len(predict)):
        print('class:{:10} probability:{:.3}'.format(class_dict[str(i)], predict[i]))
 
 
if __name__ == '__main__':
    main()

Result: