Week5 运动鞋识别

最新推荐文章于 2024-09-27 10:11:28 发布

MOLLYSMITH

最新推荐文章于 2024-09-27 10:11:28 发布

阅读量80

点赞数

文章标签： python

本文链接：https://blog.csdn.net/MOLLYSMITH/article/details/133620628

版权

Week5，运动鞋识别

🍨 本文为🔗365天深度学习训练营中的学习记录博客
🍦 参考文章：Pytorch实战 | 第P5周：运动鞋识别
🍖 原作者：K同学啊 | 接辅导、项目定制
🚀 文章来源：K同学的学习圈子

本次采用的数据集为网络上下载的天气图像数据集，。
本人电脑配置
Python 3.8.0
Pytorch 1.8.1
torchvision 1.8.1+ cuda10.2

前期准备

1. 设置GPU/CPU

本次是在gpu上对网络进行训练和测试，先识别设备，判断设备类型。

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

2. 导入数据

下载数据到主目录的文件夹的P5data文件夹里，此次数据中的训练集和测试集是直接确定好的，因此不需要进行训练集和测试集的划分。共有两个类别分别为nike和addidas，训练集中样本数量分别为251和251，测试集中样本数量分别为38和38。

import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets
import os, PIL, random, pathlib
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
data_dir = './P5data/'
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob('*'))
classNames = [str(path).split("\\") for path in data_paths]
print(classNames)
dataset_transforms = transforms.Compose([
    transforms.Resize([224, 224]),
    #transforms.RandomHorizontalFlip(),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std = [0.229, 0.224, 0.225])
])
train_dataset = datasets.ImageFolder("./P5data/train", transform=dataset_transforms)
test_dataset = datasets.ImageFolder("./P5data/test", transform = dataset_transforms)
batch_size = 32
train_dl = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle = True)
test_dl = torch.utils.data.DataLoader(test_dataset, batch_size=batch_size, shuffle = True)
for X, y in test_dl:
    print("shape of test X [N, C, H, W]:", X.shape)
    print("shape of test y:", y.shape, y.dtype)
    break
for X, y in train_dl:
    print("shape of train X [N, C, H, W]:", X.shape)
    print("shape of train y:", y.shape, y.dtype)
    break

构建网络模型

1. 搭建模型

构建含有四层卷积的卷积神经网络，其中用到了batch normalization的操作，BN层一般位于激活函数前，可以保证激活单元的非线性表达能力，缓解梯度消失问题。同时在分类器中利用了dropout，概率为0.2.

import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1=nn.Sequential(
            nn.Conv2d(3, 12, kernel_size=5, padding=0), # 12*220*220
            nn.BatchNorm2d(12),
            nn.ReLU())
        
        self.conv2=nn.Sequential(
            nn.Conv2d(12, 12, kernel_size=5, padding=0), # 12*216*216
            nn.BatchNorm2d(12),
            nn.ReLU())
        
        self.pool3=nn.Sequential(
            nn.MaxPool2d(2))                              # 12*108*108
        
        self.conv4=nn.Sequential(
            nn.Conv2d(12, 24, kernel_size=5, padding=0), # 24*104*104
            nn.BatchNorm2d(24),
            nn.ReLU())
        
        self.conv5=nn.Sequential(
            nn.Conv2d(24, 24, kernel_size=5, padding=0), # 24*100*100
            nn.BatchNorm2d(24),
            nn.ReLU())
        
        self.pool6=nn.Sequential(
            nn.MaxPool2d(2))                              # 24*50*50

        self.dropout = nn.Sequential(
            nn.Dropout(0.2))
        
        self.fc=nn.Sequential(
            nn.Linear(24*50*50, len(classeNames)))
        
    def forward(self, x):
        
        batch_size = x.size(0)
        x = self.conv1(x)  # 卷积-BN-激活
        x = self.conv2(x)  # 卷积-BN-激活
        x = self.pool3(x)  # 池化
        x = self.conv4(x)  # 卷积-BN-激活
        x = self.conv5(x)  # 卷积-BN-激活
        x = self.pool6(x)  # 池化
        x = self.dropout(x)
        x = x.view(batch_size, -1)  # flatten 变成全连接网络需要的输入 (batch, 24*50*50) ==> (batch, -1), -1 此处自动算出的是24*50*50
        x = self.fc(x)
       
        return x

device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))

model = Model().to(device)
model

3. 编写训练函数

设置损失函数，这里采用的交叉熵损失函数，设置优化器为SGD优化，同时在其中加入了动量，也是为了防止过拟合。

# 训练循环
def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)  # 训练集的大小
    num_batches = len(dataloader)   # 批次数目, (size/batch_size，向上取整)

    train_loss, train_acc = 0, 0  # 初始化训练损失和正确率
    
    for X, y in dataloader:  # 获取图片及其标签
        X, y = X.to(device), y.to(device)
        
        # 计算预测误差
        pred = model(X)          # 网络输出
        loss = loss_fn(pred, y)  # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失
        
        # 反向传播
        optimizer.zero_grad()  # grad属性归零
        loss.backward()        # 反向传播
        optimizer.step()       # 每一步自动更新
        
        # 记录acc与loss
        train_acc  += (pred.argmax(1) == y).type(torch.float).sum().item()
        train_loss += loss.item()
            
    train_acc  /= size
    train_loss /= num_batches

    return train_acc, train_loss

4. 编写测试函数

当不进行训练时，停止梯度更新，节省计算内存消耗。

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

5. 主函数

设置迭代epoch次数，这里设定为40，并记录训练误差、精度，测试误差、精度。在这里采用了学习率自适应的方式，即学习率随着epoch的增加而逐渐减小，。

def adjust_learning_rate(optimizer, epoch, start_lr):
    # 每 2 个epoch衰减到原来的 0.98
    lr = start_lr * (0.92 ** (epoch // 2))
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr

learn_rate = 1e-4 # 初始学习率
optimizer  = torch.optim.SGD(model.parameters(), lr=learn_rate)

loss_fn    = nn.CrossEntropyLoss() # 创建损失函数
epochs     = 40

train_loss = []
train_acc  = []
test_loss  = []
test_acc   = []

for epoch in range(epochs):
    # 更新学习率（使用自定义学习率时使用）
    adjust_learning_rate(optimizer, epoch, learn_rate)
    
    model.train()
    epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer)
    # scheduler.step() # 更新学习率（调用官方动态学习率接口时使用）
    
    model.eval()
    epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)
    
    train_acc.append(epoch_train_acc)
    train_loss.append(epoch_train_loss)
    test_acc.append(epoch_test_acc)
    test_loss.append(epoch_test_loss)
    
    # 获取当前的学习率
    lr = optimizer.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')

结果总结

（1）采用原始网络进行训练得到的训练结果如下
在这里插入图片描述

训练过程如下

Epoch: 1, Train_acc:46.2%, Train_loss:0.822, Test_acc:51.3%, Test_loss:0.677, Lr:1.00E-04
Epoch: 2, Train_acc:55.4%, Train_loss:0.708, Test_acc:67.1%, Test_loss:0.628, Lr:1.00E-04
Epoch: 3, Train_acc:64.7%, Train_loss:0.636, Test_acc:67.1%, Test_loss:0.577, Lr:9.20E-05
Epoch: 4, Train_acc:70.3%, Train_loss:0.574, Test_acc:73.7%, Test_loss:0.597, Lr:9.20E-05
Epoch: 5, Train_acc:73.7%, Train_loss:0.549, Test_acc:71.1%, Test_loss:0.533, Lr:8.46E-05
Epoch: 6, Train_acc:76.1%, Train_loss:0.544, Test_acc:68.4%, Test_loss:0.598, Lr:8.46E-05
Epoch: 7, Train_acc:75.1%, Train_loss:0.531, Test_acc:73.7%, Test_loss:0.542, Lr:7.79E-05
Epoch: 8, Train_acc:76.1%, Train_loss:0.515, Test_acc:67.1%, Test_loss:0.572, Lr:7.79E-05
Epoch: 9, Train_acc:79.5%, Train_loss:0.476, Test_acc:73.7%, Test_loss:0.498, Lr:7.16E-05
Epoch:10, Train_acc:80.3%, Train_loss:0.459, Test_acc:73.7%, Test_loss:0.508, Lr:7.16E-05
Epoch:11, Train_acc:84.7%, Train_loss:0.435, Test_acc:73.7%, Test_loss:0.486, Lr:6.59E-05
Epoch:12, Train_acc:82.3%, Train_loss:0.441, Test_acc:73.7%, Test_loss:0.496, Lr:6.59E-05
Epoch:13, Train_acc:86.1%, Train_loss:0.405, Test_acc:75.0%, Test_loss:0.522, Lr:6.06E-05
Epoch:14, Train_acc:87.3%, Train_loss:0.409, Test_acc:76.3%, Test_loss:0.488, Lr:6.06E-05
Epoch:15, Train_acc:87.3%, Train_loss:0.387, Test_acc:77.6%, Test_loss:0.477, Lr:5.58E-05
Epoch:16, Train_acc:88.4%, Train_loss:0.381, Test_acc:77.6%, Test_loss:0.488, Lr:5.58E-05
Epoch:17, Train_acc:89.0%, Train_loss:0.366, Test_acc:77.6%, Test_loss:0.471, Lr:5.13E-05
Epoch:18, Train_acc:88.2%, Train_loss:0.373, Test_acc:76.3%, Test_loss:0.448, Lr:5.13E-05
Epoch:19, Train_acc:88.6%, Train_loss:0.363, Test_acc:78.9%, Test_loss:0.486, Lr:4.72E-05
Epoch:20, Train_acc:89.4%, Train_loss:0.353, Test_acc:77.6%, Test_loss:0.455, Lr:4.72E-05
Epoch:21, Train_acc:90.6%, Train_loss:0.351, Test_acc:76.3%, Test_loss:0.451, Lr:4.34E-05
Epoch:22, Train_acc:89.6%, Train_loss:0.345, Test_acc:78.9%, Test_loss:0.461, Lr:4.34E-05
Epoch:23, Train_acc:91.8%, Train_loss:0.329, Test_acc:80.3%, Test_loss:0.462, Lr:4.00E-05
Epoch:24, Train_acc:91.2%, Train_loss:0.331, Test_acc:77.6%, Test_loss:0.535, Lr:4.00E-05
Epoch:25, Train_acc:92.8%, Train_loss:0.329, Test_acc:78.9%, Test_loss:0.488, Lr:3.68E-05
Epoch:26, Train_acc:91.4%, Train_loss:0.330, Test_acc:78.9%, Test_loss:0.500, Lr:3.68E-05
Epoch:27, Train_acc:92.4%, Train_loss:0.319, Test_acc:81.6%, Test_loss:0.449, Lr:3.38E-05
Epoch:28, Train_acc:92.0%, Train_loss:0.319, Test_acc:78.9%, Test_loss:0.457, Lr:3.38E-05
Epoch:30, Train_acc:92.6%, Train_loss:0.312, Test_acc:78.9%, Test_loss:0.467, Lr:3.11E-05
Epoch:31, Train_acc:92.0%, Train_loss:0.316, Test_acc:81.6%, Test_loss:0.435, Lr:2.86E-05
Epoch:32, Train_acc:92.6%, Train_loss:0.301, Test_acc:82.9%, Test_loss:0.437, Lr:2.86E-05
Epoch:33, Train_acc:93.0%, Train_loss:0.295, Test_acc:80.3%, Test_loss:0.433, Lr:2.63E-05
Epoch:34, Train_acc:92.6%, Train_loss:0.302, Test_acc:80.3%, Test_loss:0.479, Lr:2.63E-05
Epoch:35, Train_acc:93.4%, Train_loss:0.295, Test_acc:81.6%, Test_loss:0.452, Lr:2.42E-05
Epoch:36, Train_acc:93.4%, Train_loss:0.303, Test_acc:80.3%, Test_loss:0.462, Lr:2.42E-05
Epoch:37, Train_acc:93.0%, Train_loss:0.298, Test_acc:80.3%, Test_loss:0.463, Lr:2.23E-05
Epoch:38, Train_acc:93.2%, Train_loss:0.292, Test_acc:80.3%, Test_loss:0.393, Lr:2.23E-05
Epoch:39, Train_acc:94.2%, Train_loss:0.289, Test_acc:81.6%, Test_loss:0.432, Lr:2.05E-05
Epoch:40, Train_acc:95.0%, Train_loss:0.284, Test_acc:80.3%, Test_loss:0.477, Lr:2.05E-05

（2）在使用原始网络后，在第40个epoch时，测试集精度为80.3%，在第四周学习任务中，发现增加卷积层深度并不太能增加精度，而增加网络宽度可以，因此，增加通道数，具体如下所示。

# construct the NN
import torch.nn.functional as F
class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1 = nn.Sequential(nn.Conv2d(3, 12, kernel_size = 5, padding=0),#12*220*220
                                   nn.BatchNorm2d(12),
                                   nn.ReLU())
        self.conv2 = nn.Sequential(nn.Conv2d(12, 24, kernel_size = 5, padding=0),#12*216*216
                            nn.BatchNorm2d(24),
                            nn.ReLU())
        self.pool3 = nn.Sequential(nn.MaxPool2d(2))#12*108*108
        self.conv4 = nn.Sequential(nn.Conv2d(24, 48, kernel_size = 5, padding=0),#12*104*104
                    nn.BatchNorm2d(48),
                    nn.ReLU())
        self.conv5 = nn.Sequential(nn.Conv2d(48, 96, kernel_size = 5, padding=0),#12*100*100
                    nn.BatchNorm2d(96),
                    nn.ReLU())
        self.pool6 = nn.Sequential(nn.MaxPool2d(2))
        self.dropout = nn.Sequential(
            nn.Dropout(0.5))
        
        self.fc=nn.Sequential(
            nn.Linear(96*50*50, len(classNames)))

得到的训练结果如下所示
在这里插入图片描述

每个epoch的精度和损失具体值如下，最终可以达到88%的精度。

Epoch: 1, Train_acc:53.2%, Train_loss:1.490, Test_acc:55.3%, Test_loss:0.653, Lr:1.00E-04
Epoch: 2, Train_acc:53.8%, Train_loss:1.585, Test_acc:71.1%, Test_loss:0.513, Lr:1.00E-04
Epoch: 3, Train_acc:68.7%, Train_loss:0.804, Test_acc:76.3%, Test_loss:0.473, Lr:9.20E-05
Epoch: 4, Train_acc:67.3%, Train_loss:0.790, Test_acc:50.0%, Test_loss:1.654, Lr:9.20E-05
Epoch: 5, Train_acc:73.7%, Train_loss:0.651, Test_acc:52.6%, Test_loss:1.123, Lr:8.46E-05
Epoch: 6, Train_acc:84.1%, Train_loss:0.386, Test_acc:78.9%, Test_loss:0.471, Lr:8.46E-05
Epoch: 7, Train_acc:79.9%, Train_loss:0.407, Test_acc:72.4%, Test_loss:0.487, Lr:7.79E-05
Epoch: 8, Train_acc:88.6%, Train_loss:0.282, Test_acc:73.7%, Test_loss:0.476, Lr:7.79E-05
Epoch: 9, Train_acc:91.6%, Train_loss:0.258, Test_acc:75.0%, Test_loss:0.504, Lr:7.16E-05
Epoch:10, Train_acc:92.6%, Train_loss:0.227, Test_acc:84.2%, Test_loss:0.392, Lr:7.16E-05
Epoch:11, Train_acc:93.6%, Train_loss:0.220, Test_acc:76.3%, Test_loss:0.481, Lr:6.59E-05
Epoch:12, Train_acc:94.0%, Train_loss:0.219, Test_acc:72.4%, Test_loss:0.601, Lr:6.59E-05
Epoch:13, Train_acc:93.4%, Train_loss:0.216, Test_acc:82.9%, Test_loss:0.420, Lr:6.06E-05
Epoch:14, Train_acc:96.6%, Train_loss:0.182, Test_acc:85.5%, Test_loss:0.358, Lr:6.06E-05
Epoch:15, Train_acc:96.6%, Train_loss:0.173, Test_acc:84.2%, Test_loss:0.403, Lr:5.58E-05
Epoch:16, Train_acc:96.0%, Train_loss:0.174, Test_acc:85.5%, Test_loss:0.361, Lr:5.58E-05
Epoch:17, Train_acc:95.2%, Train_loss:0.171, Test_acc:84.2%, Test_loss:0.402, Lr:5.13E-05
Epoch:18, Train_acc:97.2%, Train_loss:0.160, Test_acc:85.5%, Test_loss:0.427, Lr:5.13E-05
Epoch:19, Train_acc:97.0%, Train_loss:0.156, Test_acc:86.8%, Test_loss:0.358, Lr:4.72E-05
Epoch:20, Train_acc:97.4%, Train_loss:0.152, Test_acc:84.2%, Test_loss:0.369, Lr:4.72E-05
Epoch:21, Train_acc:97.6%, Train_loss:0.144, Test_acc:85.5%, Test_loss:0.363, Lr:4.34E-05
Epoch:22, Train_acc:96.2%, Train_loss:0.153, Test_acc:85.5%, Test_loss:0.372, Lr:4.34E-05
Epoch:23, Train_acc:97.4%, Train_loss:0.147, Test_acc:86.8%, Test_loss:0.364, Lr:4.00E-05
Epoch:24, Train_acc:98.6%, Train_loss:0.132, Test_acc:86.8%, Test_loss:0.388, Lr:4.00E-05
Epoch:25, Train_acc:98.8%, Train_loss:0.129, Test_acc:88.2%, Test_loss:0.339, Lr:3.68E-05
Epoch:26, Train_acc:98.4%, Train_loss:0.126, Test_acc:85.5%, Test_loss:0.397, Lr:3.68E-05
Epoch:27, Train_acc:97.0%, Train_loss:0.129, Test_acc:86.8%, Test_loss:0.312, Lr:3.38E-05
Epoch:28, Train_acc:98.6%, Train_loss:0.116, Test_acc:85.5%, Test_loss:0.361, Lr:3.38E-05
Epoch:29, Train_acc:98.4%, Train_loss:0.118, Test_acc:86.8%, Test_loss:0.376, Lr:3.11E-05
Epoch:30, Train_acc:98.8%, Train_loss:0.128, Test_acc:86.8%, Test_loss:0.386, Lr:3.11E-05
Epoch:31, Train_acc:98.4%, Train_loss:0.120, Test_acc:88.2%, Test_loss:0.362, Lr:2.86E-05
Epoch:32, Train_acc:98.8%, Train_loss:0.113, Test_acc:88.2%, Test_loss:0.336, Lr:2.86E-05
Epoch:33, Train_acc:98.6%, Train_loss:0.118, Test_acc:85.5%, Test_loss:0.351, Lr:2.63E-05
Epoch:34, Train_acc:98.8%, Train_loss:0.113, Test_acc:86.8%, Test_loss:0.403, Lr:2.63E-05
Epoch:35, Train_acc:98.8%, Train_loss:0.116, Test_acc:86.8%, Test_loss:0.346, Lr:2.42E-05
Epoch:36, Train_acc:98.2%, Train_loss:0.112, Test_acc:88.2%, Test_loss:0.309, Lr:2.42E-05
Epoch:37, Train_acc:98.0%, Train_loss:0.120, Test_acc:88.2%, Test_loss:0.357, Lr:2.23E-05
Epoch:38, Train_acc:98.4%, Train_loss:0.107, Test_acc:88.2%, Test_loss:0.394, Lr:2.23E-05
Epoch:39, Train_acc:99.2%, Train_loss:0.104, Test_acc:88.2%, Test_loss:0.345, Lr:2.05E-05
Epoch:40, Train_acc:99.2%, Train_loss:0.106, Test_acc:88.2%, Test_loss:0.369, Lr:2.05E-05