【深度学习实战】p5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
• 🍖 原作者：K同学啊

代码

import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets

import os,PIL,pathlib

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

device

device(type=‘cuda’)

import os,PIL,random,pathlib

data_dir = './data/p5-data/'
data_dir = pathlib.Path(data_dir)

data_paths  = list(data_dir.glob('*'))
classeNames = [str(path).split("\\")[1] for path in data_paths]
classeNames

[‘p5-data’, ‘p5-data’]

# 关于transforms.Compose的更多介绍可以参考：https://blog.csdn.net/qq_38251616/article/details/124878863
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。
])

train_dataset = datasets.ImageFolder("./data/p5-data/train/",transform=train_transforms)
test_dataset  = datasets.ImageFolder("./data/p5-data/test/",transform=train_transforms)

batch_size = 32

train_dl = torch.utils.data.DataLoader(train_dataset,
                                           batch_size=batch_size,
                                           shuffle=True,
                                           num_workers=1)
test_dl = torch.utils.data.DataLoader(test_dataset,
                                          batch_size=batch_size,
                                          shuffle=True,
                                          num_workers=1)

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=3, padding=0), # 12*220*220
            nn.BatchNorm2d(12),
            nn.ReLU())
        self.conv12=nn.Sequential(
            nn.Conv2d(12, 12, kernel_size=3, 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.conv12(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

Using cuda device
Model(
(conv1): Sequential(
(0): Conv2d(3, 12, kernel_size=(3, 3), stride=(1, 1))
(1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv12): Sequential(
(0): Conv2d(12, 12, kernel_size=(3, 3), stride=(1, 1))
(1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv2): Sequential(
(0): Conv2d(12, 12, kernel_size=(5, 5), stride=(1, 1))
(1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(pool3): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(conv4): Sequential(
(0): Conv2d(12, 24, kernel_size=(5, 5), stride=(1, 1))
(1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(conv5): Sequential(
(0): Conv2d(24, 24, kernel_size=(5, 5), stride=(1, 1))
(1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
)
(pool6): Sequential(
(0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(dropout): Sequential(
(0): Dropout(p=0.2, inplace=False)
)
(fc): Sequential(
(0): Linear(in_features=60000, out_features=2, bias=True)
)
)

# 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

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

def adjust_learning_rate(optimizer, epoch, start_lr):
    # 每 2 个epoch衰减到原来的 0.92
    lr = start_lr * (0.92 ** (epoch // 7))
    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)

# learn_rate = 5e-5 # 学习率
# optimizer  = torch.optim.Adam(model.parameters(),lr=learn_rate)

loss_fn    = nn.CrossEntropyLoss() # 创建损失函数
epochs     = 60

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))

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

Epoch: 1, Train_acc:52.0%, Train_loss:0.780, Test_acc:51.3%, Test_loss:0.680, Lr:1.00E-04
Epoch: 2, Train_acc:59.2%, Train_loss:0.692, Test_acc:53.9%, Test_loss:0.669, Lr:1.00E-04
Epoch: 3, Train_acc:66.9%, Train_loss:0.630, Test_acc:52.6%, Test_loss:0.704, Lr:1.00E-04
Epoch: 4, Train_acc:71.5%, Train_loss:0.572, Test_acc:71.1%, Test_loss:0.557, Lr:1.00E-04
Epoch: 5, Train_acc:73.7%, Train_loss:0.534, Test_acc:72.4%, Test_loss:0.554, Lr:1.00E-04
Epoch: 6, Train_acc:75.9%, Train_loss:0.508, Test_acc:68.4%, Test_loss:0.657, Lr:1.00E-04
Epoch: 7, Train_acc:77.9%, Train_loss:0.477, Test_acc:72.4%, Test_loss:0.499, Lr:1.00E-04
Epoch: 8, Train_acc:82.3%, Train_loss:0.438, Test_acc:72.4%, Test_loss:0.500, Lr:9.20E-05
Epoch: 9, Train_acc:81.9%, Train_loss:0.435, Test_acc:78.9%, Test_loss:0.493, Lr:9.20E-05
Epoch:10, Train_acc:83.1%, Train_loss:0.403, Test_acc:78.9%, Test_loss:0.492, Lr:9.20E-05
Epoch:11, Train_acc:85.3%, Train_loss:0.399, Test_acc:77.6%, Test_loss:0.468, Lr:9.20E-05
Epoch:12, Train_acc:88.0%, Train_loss:0.379, Test_acc:80.3%, Test_loss:0.460, Lr:9.20E-05
Epoch:13, Train_acc:87.5%, Train_loss:0.365, Test_acc:81.6%, Test_loss:0.465, Lr:9.20E-05
Epoch:14, Train_acc:86.5%, Train_loss:0.358, Test_acc:80.3%, Test_loss:0.464, Lr:9.20E-05
Epoch:15, Train_acc:89.0%, Train_loss:0.334, Test_acc:81.6%, Test_loss:0.465, Lr:8.46E-05
Epoch:16, Train_acc:90.4%, Train_loss:0.330, Test_acc:80.3%, Test_loss:0.410, Lr:8.46E-05
Epoch:17, Train_acc:89.8%, Train_loss:0.327, Test_acc:82.9%, Test_loss:0.410, Lr:8.46E-05
Epoch:18, Train_acc:92.8%, Train_loss:0.306, Test_acc:80.3%, Test_loss:0.400, Lr:8.46E-05
Epoch:19, Train_acc:93.0%, Train_loss:0.303, Test_acc:82.9%, Test_loss:0.397, Lr:8.46E-05
Epoch:20, Train_acc:93.4%, Train_loss:0.292, Test_acc:82.9%, Test_loss:0.389, Lr:8.46E-05
Epoch:21, Train_acc:92.6%, Train_loss:0.292, Test_acc:82.9%, Test_loss:0.389, Lr:8.46E-05
Epoch:22, Train_acc:93.8%, Train_loss:0.283, Test_acc:82.9%, Test_loss:0.431, Lr:7.79E-05
Epoch:23, Train_acc:94.8%, Train_loss:0.267, Test_acc:82.9%, Test_loss:0.412, Lr:7.79E-05
Epoch:24, Train_acc:94.8%, Train_loss:0.268, Test_acc:82.9%, Test_loss:0.398, Lr:7.79E-05
Epoch:25, Train_acc:94.4%, Train_loss:0.261, Test_acc:80.3%, Test_loss:0.453, Lr:7.79E-05
Epoch:26, Train_acc:95.4%, Train_loss:0.241, Test_acc:84.2%, Test_loss:0.423, Lr:7.79E-05
Epoch:27, Train_acc:95.2%, Train_loss:0.235, Test_acc:85.5%, Test_loss:0.405, Lr:7.79E-05
Epoch:28, Train_acc:95.2%, Train_loss:0.237, Test_acc:81.6%, Test_loss:0.388, Lr:7.79E-05
Epoch:29, Train_acc:96.2%, Train_loss:0.238, Test_acc:85.5%, Test_loss:0.388, Lr:7.16E-05
Epoch:30, Train_acc:96.6%, Train_loss:0.227, Test_acc:81.6%, Test_loss:0.358, Lr:7.16E-05
Epoch:31, Train_acc:96.4%, Train_loss:0.229, Test_acc:81.6%, Test_loss:0.400, Lr:7.16E-05
Epoch:32, Train_acc:96.2%, Train_loss:0.219, Test_acc:85.5%, Test_loss:0.382, Lr:7.16E-05
Epoch:33, Train_acc:96.4%, Train_loss:0.214, Test_acc:85.5%, Test_loss:0.366, Lr:7.16E-05
Epoch:34, Train_acc:96.8%, Train_loss:0.212, Test_acc:86.8%, Test_loss:0.362, Lr:7.16E-05
Epoch:35, Train_acc:97.4%, Train_loss:0.204, Test_acc:84.2%, Test_loss:0.360, Lr:7.16E-05
Epoch:36, Train_acc:97.2%, Train_loss:0.208, Test_acc:81.6%, Test_loss:0.413, Lr:6.59E-05
Epoch:37, Train_acc:97.6%, Train_loss:0.194, Test_acc:85.5%, Test_loss:0.370, Lr:6.59E-05
Epoch:38, Train_acc:96.8%, Train_loss:0.198, Test_acc:86.8%, Test_loss:0.379, Lr:6.59E-05
Epoch:39, Train_acc:97.4%, Train_loss:0.196, Test_acc:85.5%, Test_loss:0.411, Lr:6.59E-05
Epoch:40, Train_acc:98.0%, Train_loss:0.191, Test_acc:85.5%, Test_loss:0.366, Lr:6.59E-05
Epoch:41, Train_acc:98.2%, Train_loss:0.191, Test_acc:81.6%, Test_loss:0.355, Lr:6.59E-05
Epoch:42, Train_acc:96.8%, Train_loss:0.189, Test_acc:84.2%, Test_loss:0.363, Lr:6.59E-05
Epoch:43, Train_acc:97.4%, Train_loss:0.190, Test_acc:88.2%, Test_loss:0.347, Lr:6.06E-05
Epoch:44, Train_acc:97.0%, Train_loss:0.190, Test_acc:85.5%, Test_loss:0.372, Lr:6.06E-05
Epoch:45, Train_acc:97.6%, Train_loss:0.176, Test_acc:85.5%, Test_loss:0.368, Lr:6.06E-05
Epoch:46, Train_acc:97.4%, Train_loss:0.182, Test_acc:86.8%, Test_loss:0.383, Lr:6.06E-05
Epoch:47, Train_acc:98.0%, Train_loss:0.173, Test_acc:84.2%, Test_loss:0.354, Lr:6.06E-05
Epoch:48, Train_acc:98.4%, Train_loss:0.178, Test_acc:85.5%, Test_loss:0.343, Lr:6.06E-05
Epoch:49, Train_acc:98.2%, Train_loss:0.165, Test_acc:84.2%, Test_loss:0.373, Lr:6.06E-05
Epoch:50, Train_acc:98.2%, Train_loss:0.166, Test_acc:88.2%, Test_loss:0.385, Lr:5.58E-05
Epoch:51, Train_acc:98.0%, Train_loss:0.164, Test_acc:84.2%, Test_loss:0.362, Lr:5.58E-05
Epoch:52, Train_acc:98.2%, Train_loss:0.169, Test_acc:86.8%, Test_loss:0.352, Lr:5.58E-05
Epoch:53, Train_acc:98.0%, Train_loss:0.163, Test_acc:86.8%, Test_loss:0.377, Lr:5.58E-05
Epoch:54, Train_acc:98.6%, Train_loss:0.159, Test_acc:86.8%, Test_loss:0.420, Lr:5.58E-05
Epoch:55, Train_acc:98.8%, Train_loss:0.158, Test_acc:85.5%, Test_loss:0.379, Lr:5.58E-05
Epoch:56, Train_acc:98.0%, Train_loss:0.158, Test_acc:88.2%, Test_loss:0.380, Lr:5.58E-05
Epoch:57, Train_acc:98.8%, Train_loss:0.154, Test_acc:88.2%, Test_loss:0.369, Lr:5.13E-05
Epoch:58, Train_acc:98.0%, Train_loss:0.157, Test_acc:85.5%, Test_loss:0.336, Lr:5.13E-05
Epoch:59, Train_acc:98.2%, Train_loss:0.158, Test_acc:86.8%, Test_loss:0.345, Lr:5.13E-05
Epoch:60, Train_acc:98.4%, Train_loss:0.154, Test_acc:88.2%, Test_loss:0.340, Lr:5.13E-05
Done

import matplotlib.pyplot as plt
#隐藏警告
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.axhline(y=0.88, color='r', linestyle='--', linewidth=1)
plt.axhline(y=0.9, color='g', linestyle='--', linewidth=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()

# 模型保存
PATH = './p5_model.pth'  # 保存的参数文件名
torch.save(model.state_dict(), PATH)

# 将参数加载到model当中
model.load_state_dict(torch.load(PATH, map_location=device))

from PIL import Image 

classes = list(train_dataset.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='./data/p5-data/test/adidas/1.jpg', 
                  model=model, 
                  transform=train_transforms, 
                  classes=classes)

预测结果是：adidas

思考问题

1. 卷积核大小对于训练得影响
   李沐得动手学视频中在说VGG网络时，说同样参属下，堆叠更多得3*3要比5*5得效果要好，个人感觉，使用3*3得卷积核，肯定层数更深，所以效果好，但是由于感受也变小了，所以有可能会有些较大得特征并不能识别到，不知道会不会是由于层次更深，所以识别的效果更好，这个还有待验证，等下一节vgg16尽量做一下验证
2. padding对于训练得影响
   今天做了测试，对于卷积核5*5，加了padding=2之后得效果并不好，所以在考虑padding得问题，后续也需要做验证
3. 对于训练的随机性
   同样的参数和数据，也许是因为随机得分组不同，导致训练得结果差异很大，所以要考虑是哪些因素导致的，是数据集不够大，还是训练轮次不够多，这个也需要后续验证，本节得训练最高准确率达到88.2，属实意外
4. 对于动态学习率
   前面就思考过这个问题，学习率还是要手动设置，具体多少轮次衰减，衰减到上一次得多少，还是要经验多一点