图像分类训练方案优化设计

$$图像分类训练设计$$

# &＃61;&＃61; 引入工具包 &＃61;&＃61;
import torch
from torch import nn
import torch.nn.functional as F
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
import torch.optim as optim
# &＃61;&＃61; step 0 参数配置 &＃61;&＃61;# &＃61;&＃61; step 1 数据处理 &＃61;&＃61;
norm_mean &＃61; [0.33424968,0.33424437, 0.33428448]
norm_std &＃61; [0.24796878, 0.24796101, 0.24801227]train_transform &＃61; transforms.Compose([transforms.Resize((32, 32)),transforms.ToTensor(), # 0-255 归一化到0-1 转Tensortransforms.Normalize(norm_mean, norm_std),
])train_data_path &＃61; r"D:\PycharmProjects\AI_Easy_Demo\MyData\split_data\train"
from MyDataset.Cifar10_Dataset import LoadDataset
train_dataset &＃61; LoadDataset(data_dir&＃61;r"D:\PycharmProjects\AI_Easy_Demo\MyData\split_data\train",transform&＃61;train_transform)
train_loader &＃61; DataLoader(dataset&＃61;train_dataset, batch_size&＃61;8, shuffle&＃61;True) # shuffle训练时打乱样本
# &＃61;&＃61; step 2 模型 &＃61;&＃61;
from MyNet.ResNet import ResNet34
net &＃61; ResNet34(num_classes&＃61;10, num_linear&＃61;512)# &＃61;&＃61; step 3 损失函数 &＃61;&＃61;
criterion &＃61; nn.NLLLoss()# &＃61;&＃61; step 4 优化器 &＃61;&＃61;
optimizer &＃61; optim.SGD(net.parameters(), lr&＃61;0.01, momentum&＃61;0.9) # 选择优化器
scheduler &＃61; torch.optim.lr_scheduler.StepLR(optimizer, step_size&＃61;10, gamma&＃61;0.1) # 设置学习率下降策略&＃xff0c;每过step_size个epoch&＃xff0c;做一次更新
# &＃61;&＃61; step 5 评测函数&＃61;&＃61;# &＃61;&＃61; step 6 训练 &＃61;&＃61;
MAX_EPOCH&＃61;200
best &＃61; [0] # 存储最优指标&＃xff0c;用于Early Stoppingfor i in range(MAX_EPOCH):print("当前轮转次数:",i&＃43;1)for idx, data_info in enumerate(train_loader):# print("训练数据索引",idx)inputs, labels &＃61; data_infooutputs &＃61; net(inputs)outputs &＃61; F.log_softmax(outputs, dim&＃61;1)optimizer.zero_grad() # 梯度置零,设置在loss之前loss &＃61; criterion(outputs,labels)loss.backward() # loss反向传播&＃xff0c;梯度的计算print(loss)# update weightsoptimizer.step() # 更新所有的参数 根据误差和梯度进行权重的更新# if best_loss torch.save(net.state_dict(), "best.pth")scheduler.step() # 更新学习率# &＃61;&＃61; step 7 训练可视化 &＃61;&＃61;# &＃61;&＃61; inference &＃61;&＃61;