LeNet模型训练与预测

LeNet模型训练与预测

预测的classes = (‘plane’, ‘car’, ‘bird’, ‘cat’, ‘deer’, ‘dog’, ‘frog’, ‘horse’, ‘ship’, ‘truck’)

predict.py

import torch
import torchvision.transforms as transforms
from model import LeNet
import torch
from PIL import Image
from torchvision import transforms
import matplotlib.pyplot as plt
import cv2

def main():
    transform = transforms.Compose(
        [transforms.Resize((32, 32)), #缩放
         transforms.ToTensor(),
         transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])

    classes = ('plane', 'car', 'bird', 'cat',
               'deer', 'dog', 'frog', 'horse', 'ship', 'truck')

    net = LeNet()
    net.load_state_dict(torch.load('Lenet.pth'))

    im = Image.open('data\\cifar-10-batches-py\\NO.45class7horse.jpg')
    img=cv2.imread('data\\cifar-10-batches-py\\NO.45class7horse.jpg')

    im = transform(im)  # [C, H, W]
    im = torch.unsqueeze(im, dim=0)  # 在转换为 Python Tensor 的通道排序:[N, C, H, W]; dim=0 在最前面增加维度

    with torch.no_grad(): #不求损失梯度，而网络默认会求损失梯度
        outputs = net(im)
        predict = torch.max(outputs, dim=1)[1].numpy()
    print(classes[int(predict)])#把索引index 传给classes，得到分类
    pre =torch.softmax(outputs,dim=1),#再次也可用softmax替代max来进行分类。
    print(pre)
    plt.title(classes[int(predict)])
    data = cv2.resize(img,dsize=None,fx=5,fy=5,interpolation=cv2.INTER_LINEAR)
    cv2.putText(data, classes[int(predict)], (0,40), cv2.FONT_HERSHEY_COMPLEX, 1,  (0, 255, 0), 1)
    cv2.imshow('1',data)
    cv2.waitKey(0)


if __name__ == '__main__':
    main()

model.py

import torch.nn as nn
import torch.nn.functional as F


class LeNet(nn.Module):#继承nn.Module
    def __init__(self):
        super(LeNet, self).__init__()
        self.conv1 = nn.Conv2d(3, 16, 5)#3通道的输入，16卷积核，5x5的卷积核
        self.pool1 = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(16, 32, 5)
        self.pool2 = nn.MaxPool2d(2, 2)
        self.fc1 = nn.Linear(32*5*5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)  #用的CIFAR10数据集

    def forward(self, x):#正向传播的过程
        x = F.relu(self.conv1(x))    # input(3, 32, 32) output(16, 28, 28)
        x = self.pool1(x)            # output(16, 14, 14)
        x = F.relu(self.conv2(x))    # output(32, 10, 10)
        x = self.pool2(x)            # output(32, 5, 5)
        x = x.view(-1, 32*5*5)       # output(32*5*5)    #View函数展平操作，-1代表纬度，自动推理纬度，32*5*5展平后节点个数
        x = F.relu(self.fc1(x))      # output(120)
        x = F.relu(self.fc2(x))      # output(84)
        x = self.fc3(x)              # output(10)   #最后一层，这里不需要添加softmax层了，train.py在卷网络中，在卷积交叉熵中nn.CrossEntropyLoss()，他的内部实现了分类的功能
        return x

train.py

import torch
import torchvision
import torch.nn as nn
from model import LeNet
import torch.optim as optim
import torchvision.transforms as transforms
import numpy as np
import matplotlib.pyplot as plt

def main():
    transform = transforms.Compose(     #预处理方法打包
        [transforms.ToTensor(),   #[0, 255] to a torch.FloatTensor of shape (C x H x W) in the range [0.0, 1.0]
         transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])  #``output[channel] = (input[channel] - mean[channel]) / std[channel]``

    # 50000张训练图片
    # 第一次使用时要将download设置为True才会自动去下载数据集
    train_set = torchvision.datasets.CIFAR10(root='./data', train=True,
                                             download=True, transform=transform)


                                             
    train_loader = torch.utils.data.DataLoader(train_set, batch_size=36,
                                               shuffle=True, num_workers=0)  # shuffle=True打乱数据 ，num_workers=0载入数据线程数，windos下只能设置 为0，linuxx下可设置值

    # 10000张验证图片；   train=False测试集
    # 第一次使用时要将download设置为True才会自动去下载数据集
    val_set = torchvision.datasets.CIFAR10(root='./data', train=False,
                                           download=False, transform=transform)



    val_loader = torch.utils.data.DataLoader(val_set, batch_size=10000,
                                             shuffle=False, num_workers=0) #每次载入 batch_size=5000张
    val_data_iter = iter(val_loader)#iter 迭代器
    val_image, val_label = val_data_iter.next()
    
    classes = ('plane', 'car', 'bird', 'cat','deer', 'dog', 'frog', 'horse', 'ship', 'truck')

    #查看下数据集，训练用不到
    # def imShow(img):
    #    img = img/2+0.5
    #    npIm = img.numpy()
    #    plt.imshow(np.transpose(npIm,(1,2,0)))  #[channel,height,width] 还原成载入图像的shape[height,width,channel]
    #    plt.show()
    #    #img = torchvision.utils.make_grida(img)
    # imShow( torchvision.utils.make_grid(val_image))


    net = LeNet()                      #实例化模型
    loss_function = nn.CrossEntropyLoss()      #定义损失函数This criterion combines :class:`~torch.nn.LogSoftmax` and :class:`~torch.nn.NLLLoss` in one single class.
    optimizer = optim.Adam(net.parameters(), lr=0.001)  #net.parameters() 就是LeNet所需要训练的参数

    for epoch in range(5):  # loop over the dataset multiple times

        running_loss = 0.0 #累加训练过程的损失
        for step, data in enumerate(train_loader, start=0): #循环-->遍历训练集样本
            # get the inputs; data is a list of [inputs, labels]
            inputs, labels = data

            # zero the parameter gradients
            optimizer.zero_grad() #将历史损失梯度清零，如果不清除历史梯度，就会对计算历史梯度进行累加（思维转变：相当于我们用了一个很大的batchSize，让效果更好）
            # forward + backward + optimize
            outputs = net(inputs)
            loss = loss_function(outputs, labels)
            loss.backward() #loss反向传播
            optimizer.step()  #step() 实现参数更新

            # print statistics打印过程
            running_loss += loss.item()
            if step % 500 == 499:    # print every 500 mini-batches，每隔500步打印一次
                with torch.no_grad():  #torch.no_grad() 意思：不计算每一个节点的损失梯度，如果不加这个函数，在测试过程中也会计算损失梯度（会消耗算力，资源，会存储每个节点的损失梯度，消耗内存，容易导致程序内存不足而崩溃）
                    outputs = net(val_image)  # [batch, 10]  #正向传播
                    predict_y = torch.max(outputs, dim=1)[1]  #max网络预测类别归属（ dim=1纬度1上寻找最大值，10个节点中寻找最大值），[1] 代表我们只需要index即索引值
                    accuracy = torch.eq(predict_y, val_label).sum().item() / val_label.size(0)#.eq(predict_y, val_label).sum()累积预测对的数目是个terson张量； .item()将张量转换为标量; /val_label.size(0)除以样本总量

                    print('[%d, %5d] train_loss: %.3f  test_accuracy: %.3f' %
                          (epoch + 1, step + 1, running_loss / 500, accuracy))  #running_loss / 500是500步的平均训练误差
                    running_loss = 0.0  #将累积训练误差清零，再进行下500次的训练，重新累计

    print('Finished Training')

    save_path = './Lenet.pth'
    torch.save(net.state_dict(), save_path)  #模型权重参数保存net.state_dict()


if __name__ == '__main__':
    main()


#train_set = torchvision.datasets. 官方在这下面为我们提供了大量数据集，有时间.出来