ai-trash-can/train_mobilenetv2.py

import time
from tqdm import tqdm

import torch
import torch.nn as nn
import torch.optim as optim
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import torchvision.models as models
from torch.optim.lr_scheduler import ReduceLROnPlateau

if __name__=="__main__":
    start_time = time.time()
    # 设置随机种子，以确保结果可重复
    torch.manual_seed(114514)

    # 设置设备
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

    # 数据增强和标准化
    transform = transforms.Compose([
        transforms.RandomResizedCrop(224),
        transforms.RandomHorizontalFlip(),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ])

    # 数据加载
    train_dir = './train_data/1/train'
    test_dir = 'train_data/1/test'

    # 训练轮数
    num_epochs = 20

    #加载数据集
    train_dataset = datasets.ImageFolder(train_dir, transform=transform)
    train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=64, shuffle=True)
    test_dataset = datasets.ImageFolder(test_dir, transform=transform)
    test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=64, shuffle=False)

    # 构建MobileNetV2模型
    model = models.mobilenet_v2(pretrained=True)

    num_classes = len(train_dataset.classes)
    model.classifier[1] = nn.Linear(in_features=1280, out_features=num_classes)

    # 将模型移动到设备上
    model = model.to(device)

    # 定义损失函数和优化器
    criterion = nn.CrossEntropyLoss()
    optimizer = optim.Adam(model.parameters(), lr=0.0001)

    # 添加ReduceLROnPlateau调度器
    scheduler = ReduceLROnPlateau(optimizer, mode='min', factor=0.1, patience=2, verbose=False)

    print(f"train data:{len(train_loader)}")
    print(f"test data:{len(test_loader)}")
    print(f"epochs:{num_epochs}")

    # 训练模型
    print("start training")

    min_loss = 10000.0
    max_accuracy = 0
    #temp = time.time()
    for epoch in range(num_epochs):
        train_start_time = time.time()
        print(f"turn {epoch + 1}:")
        current_lr = optimizer.param_groups[0]['lr']
        print(f"Current learning rate: {current_lr}")
        model.train()
        running_loss = 0.0
        for inputs, labels in tqdm(train_loader, desc="training", unit="item", ncols=100):
            inputs, labels = inputs.to(device), labels.to(device)

            optimizer.zero_grad()
            outputs = model(inputs)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()

            running_loss += loss.item()

        train_end_time = time.time()
        print(f"Epoch {epoch + 1}/{num_epochs}    Train loss: {running_loss / len(train_loader)}    Train cost:{train_end_time -train_start_time}")

        # 在测试集上评估模型
        test_start_time = time.time()
        model.eval()
        correct = 0
        total = 0
        val_loss = 0.0
        with torch.no_grad():
            for inputs, labels in test_loader:
                inputs, labels = inputs.to(device), labels.to(device)

                outputs = model(inputs)
                loss = criterion(outputs, labels)
                val_loss += loss.item()

                _, predicted = torch.max(outputs.data, 1)
                total += labels.size(0)
                correct += (predicted == labels).sum().item()


        val_loss /= len(test_loader)

        # 更新学习率
        scheduler.step(val_loss)

        test_end_time = time.time()
        accuracy = 100 * correct / total
        print(f"Test Accuracy: {accuracy:.2f}%    Test loss:{ val_loss }    test cost:{ test_end_time - test_start_time }")

        # 保存模型
        if val_loss < min_loss or max_accuracy < accuracy:
            min_loss = val_loss
            max_accuracy = accuracy
            torch.save(model.state_dict(), f"./model/1/epochs{epoch + 1} {accuracy:.2f}.pt")
            print("model saved")

    print("all finish")

    '''
    torch.save(model.state_dict(), f"./model/1/epochs{epoch + 1} {accuracy:.2f}.pt")
    print("final model save")
    '''



    print(f"time:{time.time()-start_time}")