zjsh_classification/class2/train_best_pt.py

import argparse
import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import datasets, transforms
from torch.utils.data import DataLoader
from class2.test.mobilenetv3 import MobileNetV3_Large
import os

def train(args):
    # 创建输出目录
    os.makedirs("checkpoints", exist_ok=True)

    # 数据增强 & 预处理
    transform_train = 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])
    ])
    transform_val = transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406],
                             std=[0.229, 0.224, 0.225])
    ])

    # 加载数据集
    train_dataset = datasets.ImageFolder(root=f"{args.data}/train", transform=transform_train)
    val_dataset = datasets.ImageFolder(root=f"{args.data}/val", transform=transform_val)

    train_loader = DataLoader(train_dataset, batch_size=args.batch_size, shuffle=True, num_workers=4, pin_memory=True)
    val_loader = DataLoader(val_dataset, batch_size=args.batch_size, shuffle=False, num_workers=4, pin_memory=True)

    # 模型
    model = MobileNetV3_Large(num_classes=2)
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model = model.to(device)

    # 损失函数 & 优化器
    criterion = nn.CrossEntropyLoss()
    optimizer = optim.Adam(model.parameters(), lr=args.lr)

    # 学习率调度器
    scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=30, gamma=0.1)

    # ========== 改进的早停 + 保存机制 ==========
    best_val_acc = 0.0
    best_epoch = 0
    patience = args.patience  # 默认 5
    wait = 0  # 连续未超越最佳的次数

    print(f"开始训练，共 {args.epochs} 轮，早停耐心值: {patience}")

    for epoch in range(args.epochs):
        # 训练阶段
        model.train()
        running_loss = 0.0
        correct, total = 0, 0

        for images, labels in train_loader:
            images, labels = images.to(device), labels.to(device)
            optimizer.zero_grad()
            outputs = model(images)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()
            running_loss += loss.item()
            _, predicted = outputs.max(1)
            total += labels.size(0)
            correct += predicted.eq(labels).sum().item()

        train_acc = 100. * correct / total
        avg_loss = running_loss / len(train_loader)
        scheduler.step()

        # 验证阶段
        model.eval()
        correct, total = 0, 0
        with torch.no_grad():
            for images, labels in val_loader:
                images, labels = images.to(device), labels.to(device)
                outputs = model(images)
                _, predicted = outputs.max(1)
                total += labels.size(0)
                correct += predicted.eq(labels).sum().item()

        val_acc = 100. * correct / total

        print(f"[Epoch {epoch+1:2d}/{args.epochs}] "
              f"Loss: {avg_loss:.4f} | "
              f"Train Acc: {train_acc:.2f}% | "
              f"Val Acc: {val_acc:.2f}% | "
              f"LR: {scheduler.get_last_lr()[0]:.6f}")

        # ========== 核心逻辑：保存最后一个最高权重 + 改进早停 ==========
        if val_acc >= best_val_acc:  # ✅ 只要不低于最佳，就视为“保持高水平”
            if val_acc > best_val_acc:
                print(f"新的最高验证准确率: {val_acc:.2f}% (原: {best_val_acc:.2f}%)")
            else:
                print(f"验证准确率持平历史最佳: {val_acc:.2f}%")

            # 更新最佳
            best_val_acc = val_acc
            best_epoch = epoch + 1
            wait = 0  # ✅ 重置等待计数

            # 保存完整检查点（我们想要“最后一个”最高权重）
            checkpoint = {
                'epoch': epoch,
                'model_state_dict': model.state_dict(),
                'optimizer_state_dict': optimizer.state_dict(),
                'scheduler_state_dict': scheduler.state_dict(),
                'val_acc': val_acc,
                'best_val_acc': best_val_acc,
                'args': args
            }
            torch.save(checkpoint, "checkpoints/mobilenetv3_last_best.pth")
            print(f"✅ 已保存：当前最佳权重（epoch {best_epoch}）")

        else:
            # ✅ 只有严格低于最佳才增加 wait
            wait += 1
            print(f"验证准确率下降 ({val_acc:.2f}% < {best_val_acc:.2f}%)，已连续 {wait}/{patience} 次未超越")

            if wait >= patience:
                print(f"早停触发！连续 {patience} 次验证准确率未达到或超越历史最佳。")
                break

    # 训练结束
    print(f"\n✅ 训练完成！")
    print(f"最终最佳验证准确率: {best_val_acc:.2f}% (第 {best_epoch} 轮)")
    print(f"最后一个最高权重已保存至: 'checkpoints/mobilenetv3_last_best.pth'")

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--data", type=str, default="/media/hx/04e879fa-d697-4b02-ac7e-a4148876ebb0/dataset/classdata", help="数据集路径")
    parser.add_argument("--epochs", type=int, default=100, help="训练轮数")
    parser.add_argument("--batch_size", type=int, default=32, help="批大小")
    parser.add_argument("--lr", type=float, default=1e-3, help="学习率")
    parser.add_argument("--patience", type=int, default=5, help="早停耐心值（连续多少次未达到或超越历史最佳才停止）")
    args = parser.parse_args()
    train(args)