深度学习笔记（11）

官方教程图像分类这一章内容很多，我们拆成三部分来看

这一部分引入新的数据集眼疾识别数据集iChallenge-PM，并复现AlexNet来进行识别。同样也是paddle和torch对比实现。

iChallenge-PM数据集

iChallenge-PM中既有病理性近视患者的眼底图片，也有非病理性近视患者的图片，命名规则如下：

• 病理性近视（PM）：文件名以P开头
• 非病理性近视（non-PM）：
  • 高度近视（high myopia）：文件名以H开头
  • 正常眼睛（normal）：文件名以N开头

同样我们也需要写一个data_loader。这套数据与之前的mnist（以json格式存储）不一样，是以图像存储的，而label在文件名中。

import cv2
import random
import numpy as np
import os

# 对读入的图像数据进行预处理
def transform_img(img):
    # 将图片尺寸缩放道 224x224
    img = cv2.resize(img, (224, 224))
    # 读入的图像数据格式是[H, W, C]
    # 使用转置操作将其变成[C, H, W]
    img = np.transpose(img, (2,0,1))
    img = img.astype('float32')
    # 将数据范围调整到[-1.0, 1.0]之间
    img = img / 255.
    img = img * 2.0 - 1.0
    return img

# 定义训练集数据读取器
def data_loader(datadir, batch_size=10, mode = 'train'):
    # 将datadir目录下的文件列出来，每条文件都要读入
    filenames = os.listdir(datadir)
    def reader():
        if mode == 'train':
            # 训练时随机打乱数据顺序
            random.shuffle(filenames)
        batch_imgs = []
        batch_labels = []
        for name in filenames:
            filepath = os.path.join(datadir, name)
            img = cv2.imread(filepath)
            img = transform_img(img)
            if name[0] == 'H' or name[0] == 'N':
                # H开头的文件名表示高度近似，N开头的文件名表示正常视力
                # 高度近视和正常视力的样本，都不是病理性的，属于负样本，标签为0
                label = 0
            elif name[0] == 'P':
                # P开头的是病理性近视，属于正样本，标签为1
                label = 1
            else:
                raise('Not excepted file name')
            # 每读取一个样本的数据，就将其放入数据列表中
            batch_imgs.append(img)
            batch_labels.append(label)
            if len(batch_imgs) == batch_size:
                # 当数据列表的长度等于batch_size的时候，
                # 把这些数据当作一个mini-batch，并作为数据生成器的一个输出
                imgs_array = np.array(batch_imgs).astype('float32')
                labels_array = np.array(batch_labels).astype('float32').reshape(-1, 1)
                yield imgs_array, labels_array
                batch_imgs = []
                batch_labels = []

        if len(batch_imgs) > 0:
            # 剩余样本数目不足一个batch_size的数据，一起打包成一个mini-batch
            imgs_array = np.array(batch_imgs).astype('float32')
            labels_array = np.array(batch_labels).astype('float32').reshape(-1, 1)
            yield imgs_array, labels_array

    return reader

# 定义验证集数据读取器
def valid_data_loader(datadir, csvfile, batch_size=10, mode='valid'):
    # 训练集读取时通过文件名来确定样本标签，验证集则通过csvfile来读取每个图片对应的标签
    # 请查看解压后的验证集标签数据，观察csvfile文件里面所包含的内容
    # csvfile文件所包含的内容格式如下，每一行代表一个样本，
    # 其中第一列是图片id，第二列是文件名，第三列是图片标签，
    # 第四列和第五列是Fovea的坐标，与分类任务无关
    # ID,imgName,Label,Fovea_X,Fovea_Y
    # 1,V0001.jpg,0,1157.74,1019.87
    # 2,V0002.jpg,1,1285.82,1080.47
    # 打开包含验证集标签的csvfile，并读入其中的内容
    filelists = open(csvfile).readlines()
    def reader():
        batch_imgs = []
        batch_labels = []
        for line in filelists[1:]:
            line = line.strip().split(',')
            name = line[1]
            label = int(line[2])
            # 根据图片文件名加载图片，并对图像数据作预处理
            filepath = os.path.join(datadir, name)
            img = cv2.imread(filepath)
            img = transform_img(img)
            # 每读取一个样本的数据，就将其放入数据列表中
            batch_imgs.append(img)
            batch_labels.append(label)
            if len(batch_imgs) == batch_size:
                # 当数据列表的长度等于batch_size的时候，
                # 把这些数据当作一个mini-batch，并作为数据生成器的一个输出
                imgs_array = np.array(batch_imgs).astype('float32')
                labels_array = np.array(batch_labels).astype('float32').reshape(-1, 1)
                yield imgs_array, labels_array
                batch_imgs = []
                batch_labels = []

        if len(batch_imgs) > 0:
            # 剩余样本数目不足一个batch_size的数据，一起打包成一个mini-batch
            imgs_array = np.array(batch_imgs).astype('float32')
            labels_array = np.array(batch_labels).astype('float32').reshape(-1, 1)
            yield imgs_array, labels_array

    return reader

# 查看数据形状
DATADIR = '/home/aistudio/work/palm/PALM-Training400/PALM-Training400'
train_loader = data_loader(DATADIR, 
                           batch_size=10, mode='train')
data_reader = train_loader()
data = next(data_reader)
data[0].shape, data[1].shape

eval_loader = data_loader(DATADIR, 
                           batch_size=10, mode='eval')
data_reader = eval_loader()
data = next(data_reader)
data[0].shape, data[1].shape

这里居然函数嵌套定义，我不是太理解，可能这就是大佬的coding习惯吧

训练过程

# -*- coding: utf-8 -*-
# LeNet 识别眼疾图片
import os
import random
import paddle
import numpy as np

DATADIR = '/home/aistudio/work/palm/PALM-Training400/PALM-Training400'
DATADIR2 = '/home/aistudio/work/palm/PALM-Validation400'
CSVFILE = '/home/aistudio/labels.csv'

# 定义训练过程
def train_pm(model, optimizer):
    # 开启0号GPU训练
    use_gpu = True
    paddle.set_device('gpu:0') if use_gpu else paddle.set_device('cpu')

    print('start training ... ')
    model.train()
    epoch_num = 5
    # 定义数据读取器，训练数据读取器和验证数据读取器
    train_loader = data_loader(DATADIR, batch_size=10, mode='train')
    valid_loader = valid_data_loader(DATADIR2, CSVFILE)
    for epoch in range(epoch_num):
        for batch_id, data in enumerate(train_loader()):
            x_data, y_data = data
            img = paddle.to_tensor(x_data)
            label = paddle.to_tensor(y_data)
            # 运行模型前向计算，得到预测值
            logits = model(img)
            loss = F.binary_cross_entropy_with_logits(logits, label)
            avg_loss = paddle.mean(loss)

            if batch_id % 10 == 0:
                print("epoch: {}, batch_id: {}, loss is: {}".format(epoch, batch_id, avg_loss.numpy()))
            # 反向传播，更新权重，清除梯度
            avg_loss.backward()
            optimizer.step()
            optimizer.clear_grad()

        model.eval()
        accuracies = []
        losses = []
        for batch_id, data in enumerate(valid_loader()):
            x_data, y_data = data
            img = paddle.to_tensor(x_data)
            label = paddle.to_tensor(y_data)
            # 运行模型前向计算，得到预测值
            logits = model(img)
            # 二分类，sigmoid计算后的结果以0.5为阈值分两个类别
            # 计算sigmoid后的预测概率，进行loss计算
            pred = F.sigmoid(logits)
            loss = F.binary_cross_entropy_with_logits(logits, label)
            # 计算预测概率小于0.5的类别
            pred2 = pred * (-1.0) + 1.0
            # 得到两个类别的预测概率，并沿第一个维度级联
            pred = paddle.concat([pred2, pred], axis=1)
            acc = paddle.metric.accuracy(pred, paddle.cast(label, dtype='int64'))

            accuracies.append(acc.numpy())
            losses.append(loss.numpy())
        print("[validation] accuracy/loss: {}/{}".format(np.mean(accuracies), np.mean(losses)))
        model.train()

        paddle.save(model.state_dict(), 'palm.pdparams')
        paddle.save(optimizer.state_dict(), 'palm.pdopt')


# 定义评估过程
def evaluation(model, params_file_path):

    # 开启0号GPU预估
    use_gpu = True
    paddle.set_device('gpu:0') if use_gpu else paddle.set_device('cpu')

    print('start evaluation .......')

    #加载模型参数
    model_state_dict = paddle.load(params_file_path)
    model.load_dict(model_state_dict)

    model.eval()
    eval_loader = data_loader(DATADIR, 
                        batch_size=10, mode='eval')

    acc_set = []
    avg_loss_set = []
    for batch_id, data in enumerate(eval_loader()):
        x_data, y_data = data
        img = paddle.to_tensor(x_data)
        label = paddle.to_tensor(y_data)
        y_data = y_data.astype(np.int64)
        label_64 = paddle.to_tensor(y_data)
        # 计算预测和精度
        prediction, acc = model(img, label_64)
        # 计算损失函数值
        loss = F.binary_cross_entropy_with_logits(prediction, label)
        avg_loss = paddle.mean(loss)
        acc_set.append(float(acc.numpy()))
        avg_loss_set.append(float(avg_loss.numpy()))
    # 求平均精度
    acc_val_mean = np.array(acc_set).mean()
    avg_loss_val_mean = np.array(avg_loss_set).mean()

    print('loss={}, acc={}'.format(avg_loss_val_mean, acc_val_mean))

定义LeNet，由于输入的不再是 单通道灰度图片，所以第一层卷积的in_channels改成3，最后三层卷积后也不是得到120×1×1的结果而是120×50×50，所以线性层的输入变成了300000，可想而知预测结果不会理想。

# -*- coding:utf-8 -*-

# 导入需要的包
import paddle
import numpy as np
from paddle.nn import Conv2D, MaxPool2D, Linear, Dropout
import paddle.nn.functional as F

# 定义 LeNet 网络结构
class LeNet(paddle.nn.Layer):
    def __init__(self, num_classes=1):
        super(LeNet, self).__init__()

        # 创建卷积和池化层块，每个卷积层使用Sigmoid激活函数，后面跟着一个2x2的池化
        self.conv1 = Conv2D(in_channels=3, out_channels=6, kernel_size=5)
        self.max_pool1 = MaxPool2D(kernel_size=2, stride=2)
        self.conv2 = Conv2D(in_channels=6, out_channels=16, kernel_size=5)
        self.max_pool2 = MaxPool2D(kernel_size=2, stride=2)
        # 创建第3个卷积层
        self.conv3 = Conv2D(in_channels=16, out_channels=120, kernel_size=4)
        # 创建全连接层，第一个全连接层的输出神经元个数为64
        self.fc1 = Linear(in_features=300000, out_features=64)
        # 第二个全连接层输出神经元个数为分类标签的类别数
        self.fc2 = Linear(in_features=64, out_features=num_classes)

    # 网络的前向计算过程
    def forward(self, x, label=None):
        x = self.conv1(x)
        x = F.sigmoid(x)
        x = self.max_pool1(x)
        x = self.conv2(x)
        x = F.sigmoid(x)
        x = self.max_pool2(x)
        x = self.conv3(x)
        x = F.sigmoid(x)
        x = paddle.reshape(x, [x.shape[0], -1])
        x = self.fc1(x)
        x = F.sigmoid(x)
        x = self.fc2(x)
        if label is not None:
            acc = paddle.metric.accuracy(input=x, label=label)
            return x, acc
        else:
            return x

# 创建模型
model = LeNet(num_classes=1)
# 启动训练过程
opt = paddle.optimizer.Momentum(learning_rate=0.001, momentum=0.9, parameters=model.parameters())
train_pm(model, optimizer=opt)
evaluation(model, params_file_path="palm.pdparams")

start training ... 
epoch: 0, batch_id: 0, loss is: [0.50568914]
epoch: 0, batch_id: 10, loss is: [0.6118419]
epoch: 0, batch_id: 20, loss is: [0.66790974]
epoch: 0, batch_id: 30, loss is: [0.68811446]
[validation] accuracy/loss: 0.4725000262260437/0.6936749219894409
epoch: 1, batch_id: 0, loss is: [0.69677556]
epoch: 1, batch_id: 10, loss is: [0.7022215]
epoch: 1, batch_id: 20, loss is: [0.68310237]
epoch: 1, batch_id: 30, loss is: [0.70623994]
[validation] accuracy/loss: 0.5275000333786011/0.6917704343795776
epoch: 2, batch_id: 0, loss is: [0.6938783]
epoch: 2, batch_id: 10, loss is: [0.693455]
epoch: 2, batch_id: 20, loss is: [0.6738912]
epoch: 2, batch_id: 30, loss is: [0.68051213]
[validation] accuracy/loss: 0.5275000333786011/0.69181889295578
epoch: 3, batch_id: 0, loss is: [0.6810262]
epoch: 3, batch_id: 10, loss is: [0.7284224]
epoch: 3, batch_id: 20, loss is: [0.6831607]
epoch: 3, batch_id: 30, loss is: [0.7108837]
[validation] accuracy/loss: 0.5275000333786011/0.691852867603302
epoch: 4, batch_id: 0, loss is: [0.69604385]
epoch: 4, batch_id: 10, loss is: [0.69544876]
epoch: 4, batch_id: 20, loss is: [0.7562486]
epoch: 4, batch_id: 30, loss is: [0.69045544]
[validation] accuracy/loss: 0.5275000333786011/0.6916558146476746
start evaluation .......
loss=0.6911752998828888, acc=0.4675000052899122

通过运行结果可以看出，在眼疾筛查数据集iChallenge-PM上，LeNet的loss很难下降，模型没有收敛。这是因为MNIST数据集的图片尺寸比较小（28×2828\times2828×28），但是眼疾筛查数据集图片尺寸比较大（原始图片尺寸约为2000×20002000 \times 20002000×2000，经过缩放之后变成224×224224 \times 224224×224），LeNet模型很难进行有效分类。这说明在图片尺寸比较大时，LeNet在图像分类任务上存在局限性。

AlexNet

通过上面的实际训练可以看到，虽然LeNet在手写数字识别数据集上取得了很好的结果，但在更大的数据集上表现却并不好。自从1998年LeNet问世以来，接下来十几年的时间里，神经网络并没有在计算机视觉领域取得很好的结果，反而一度被其它算法所超越。原因主要有两方面，一是神经网络的计算比较复杂，对当时计算机的算力来说，训练神经网络是件非常耗时的事情；另一方面，当时还没有专门针对神经网络做算法和训练技巧的优化，神经网络的收敛是件非常困难的事情。

随着技术的进步和发展，计算机的算力越来越强大，尤其是在GPU并行计算能力的推动下，复杂神经网络的计算也变得更加容易实施。另一方面，互联网上涌现出越来越多的数据，极大的丰富了数据库。同时也有越来越多的研究人员开始专门针对神经网络做算法和模型的优化，Alex Krizhevsky等人提出的AlexNet以很大优势获得了2012年ImageNet比赛的冠军。这一成果极大的激发了产业界对神经网络的兴趣，开创了使用深度神经网络解决图像问题的途径，随后也在这一领域涌现出越来越多的优秀成果。

AlexNet与LeNet相比，具有更深的网络结构，包含5层卷积和3层全连接，同时使用了如下三种方法改进模型的训练过程：

• 数据增广：深度学习中常用的一种处理方式，通过对训练随机加一些变化，比如平移、缩放、裁剪、旋转、翻转或者增减亮度等，产生一系列跟原始图片相似但又不完全相同的样本，从而扩大训练数据集。通过这种方式，可以随机改变训练样本，避免模型过度依赖于某些属性，能从一定程度上抑制过拟合。
• 使用Dropout抑制过拟合。
• 使用ReLU激活函数减少梯度消失现象。

# -*- coding:utf-8 -*-

# 导入需要的包
import paddle
import numpy as np
from paddle.nn import Conv2D, MaxPool2D, Linear, Dropout
## 组网
import paddle.nn.functional as F

# 定义 AlexNet 网络结构
class AlexNet(paddle.nn.Layer):
    def __init__(self, num_classes=1):
        super(AlexNet, self).__init__()
        # AlexNet与LeNet一样也会同时使用卷积和池化层提取图像特征
        # 与LeNet不同的是激活函数换成了‘relu’
        self.conv1 = Conv2D(in_channels=3, out_channels=96, kernel_size=11, stride=4, padding=5)
        self.max_pool1 = MaxPool2D(kernel_size=2, stride=2)
        self.conv2 = Conv2D(in_channels=96, out_channels=256, kernel_size=5, stride=1, padding=2)
        self.max_pool2 = MaxPool2D(kernel_size=2, stride=2)
        self.conv3 = Conv2D(in_channels=256, out_channels=384, kernel_size=3, stride=1, padding=1)
        self.conv4 = Conv2D(in_channels=384, out_channels=384, kernel_size=3, stride=1, padding=1)
        self.conv5 = Conv2D(in_channels=384, out_channels=256, kernel_size=3, stride=1, padding=1)
        self.max_pool5 = MaxPool2D(kernel_size=2, stride=2)

        self.fc1 = Linear(in_features=12544, out_features=4096)
        self.drop_ratio1 = 0.5
        self.drop1 = Dropout(self.drop_ratio1)
        self.fc2 = Linear(in_features=4096, out_features=4096)
        self.drop_ratio2 = 0.5
        self.drop2 = Dropout(self.drop_ratio2)
        self.fc3 = Linear(in_features=4096, out_features=num_classes)
    
    def forward(self, x):
        x = self.conv1(x)
        x = F.relu(x)
        x = self.max_pool1(x)
        x = self.conv2(x)
        x = F.relu(x)
        x = self.max_pool2(x)
        x = self.conv3(x)
        x = F.relu(x)
        x = self.conv4(x)
        x = F.relu(x)
        x = self.conv5(x)
        x = F.relu(x)
        x = self.max_pool5(x)
        x = paddle.reshape(x, [x.shape[0], -1])
        x = self.fc1(x)
        x = F.relu(x)
        # 在全连接之后使用dropout抑制过拟合
        x = self.drop1(x)
        x = self.fc2(x)
        x = F.relu(x)
        # 在全连接之后使用dropout抑制过拟合
        x = self.drop2(x)
        x = self.fc3(x)
        return x

# 创建模型
model = AlexNet()
# 启动训练过程
opt = paddle.optimizer.Adam(learning_rate=0.001, parameters=model.parameters())

train_pm(model, optimizer=opt)

pytorch实现

改写成pytorch版本时遇到如下几个问题

• 函数接口名称不一致
• 调用gpu方式不一致
• accuracy计算方式不一样
• 数据读取报错

前三个问题前两章都提及了，不过这次的acc我使用了round和torch.eq函数来实现更为方便，因为分类结果就是0和1

在每个epoch最后计算acc时发生了数据加载函数的报错，按道理使用paddle和pytorch来实现和数据读取函数是没有关系的，为什么会报错呢？检查一下发现是csv文件最后一行出现了错误，这种事情不是第一次见到了，以后输出csv的时候都要检查最后一行是不是多了一行回车或者多了逗号等等。

另外因为我的显卡是3060laptop所以要把batchsize调小一点，要不然爆显存了

发现训练时有时会卡在下面这个acc，应该是困在某个局部最优了，重新训练一下才行

[validation] accuracy/loss: 0.5275000333786011/0.6918381452560425

最后评估过程，torch的model并不能直接输出acc，所以跟eval过程一样计算acc。

//数据读取函数同paddle

# -*- coding: utf-8 -*-
# 导入需要的包
import torch
import numpy as np
from torch.nn import Conv2d, MaxPool2d, Linear，Dropout
from torch import nn,optim
import torch.nn.functional as F
import os
import random
import numpy as np

DATADIR = r'E:\DL_DATA\iChallenge-PM\PALM-Training400\PALM-Training400'
DATADIR2 = r'E:\DL_DATA\iChallenge-PM\PALM-Validation400'
CSVFILE = r'E:\DL_DATA\iChallenge-PM\labels.csv'

# 定义训练过程
def train_pm(model, optimizer):
    print('start training ... ')
    model.cuda()
    model.train(mode=True)
    epoch_num = 5
    # 定义数据读取器，训练数据读取器和验证数据读取器
    train_loader = data_loader(DATADIR, batch_size=6,mode='train')
    valid_loader = valid_data_loader(DATADIR2, CSVFILE)
    for epoch in range(epoch_num):
        for batch_id, data in enumerate(train_loader()):
            x_data, y_data = data
            img = torch.tensor(x_data).cuda()
            label = torch.tensor(y_data).cuda()
            # 运行模型前向计算，得到预测值
            logits = model(img)
            loss = F.binary_cross_entropy_with_logits(logits, label).cuda()
            avg_loss = torch.mean(loss)

            if batch_id % 10 == 0:
                print("epoch: {}, batch_id: {}, loss is: {}".format(epoch, batch_id, avg_loss.cpu().detach().numpy()))
            # 反向传播，更新权重，清除梯度
            avg_loss.backward()
            optimizer.step()
            optimizer.zero_grad()

        model.eval()
        accuracies = []
        losses = []
        for batch_id, data in enumerate(valid_loader()):
            x_data, y_data = data
            img = torch.tensor(x_data).cuda()
            label = torch.tensor(y_data).cuda()
            # 运行模型前向计算，得到预测值
            logits = model(img)
            # 二分类，sigmoid计算后的结果以0.5为阈值分两个类别
            # 计算sigmoid后的预测概率，进行loss计算
            pred = torch.sigmoid(logits)
            loss = F.binary_cross_entropy_with_logits(logits, label)
            pred_label=pred.round().squeeze(dim=-1)
            acc=torch.eq(pred_label,label.squeeze(dim=-1)).float().mean()
            accuracies.append(acc.cpu().detach().numpy())
            losses.append(loss.cpu().detach().numpy())
        print("[validation] accuracy/loss: {}/{}".format(np.mean(accuracies), np.mean(losses)))
        model.train()
        torch.save(model.state_dict(), 'AlexNet.pdparams')
        torch.cuda.empty_cache()

# 定义评估过程
def evaluation(model, params_file_path):

    # 开启0号GPU预估
    model.cuda()
    print('start evaluation .......')

    #加载模型参数
    model.load_state_dict(torch.load(params_file_path))

    model.eval()
    eval_loader = data_loader(DATADIR, 
                        batch_size=10, mode='eval')
    
    acc_set = []
    avg_loss_set = []
    for batch_id, data in enumerate(eval_loader()):
        x_data, y_data = data
        img = torch.tensor(x_data).cuda()
        label = torch.tensor(y_data).cuda()
        y_data = y_data.astype(np.int64)
        label_64 = torch.tensor(y_data)
        # 计算预测和精度
        logits = model(img)
        # 计算损失函数值
        pred = torch.sigmoid(logits)
        loss = F.binary_cross_entropy_with_logits(logits, label)
        pred_label=pred.round().squeeze(dim=-1)
        acc=torch.eq(pred_label,label.squeeze(dim=-1)).float().mean()
        avg_loss = torch.mean(loss)
        acc_set.append(float(acc.cpu().detach().numpy()))
        avg_loss_set.append(float(avg_loss.cpu().detach().numpy()))
    # 求平均精度
    acc_val_mean = np.array(acc_set).mean()
    avg_loss_val_mean = np.array(avg_loss_set).mean()

    print('loss={}, acc={}'.format(avg_loss_val_mean, acc_val_mean))

# 定义 AlexNet 网络结构
class AlexNet(nn.Module):
    def __init__(self, num_classes=1):
        super(AlexNet, self).__init__()
        # AlexNet与LeNet一样也会同时使用卷积和池化层提取图像特征
        # 与LeNet不同的是激活函数换成了‘relu’
        self.conv1 = Conv2d(in_channels=3, out_channels=96, kernel_size=11, stride=4, padding=5)
        self.max_pool1 = MaxPool2d(kernel_size=2, stride=2)
        self.conv2 = Conv2d(in_channels=96, out_channels=256, kernel_size=5, stride=1, padding=2)
        self.max_pool2 = MaxPool2d(kernel_size=2, stride=2)
        self.conv3 = Conv2d(in_channels=256, out_channels=384, kernel_size=3, stride=1, padding=1)
        self.conv4 = Conv2d(in_channels=384, out_channels=384, kernel_size=3, stride=1, padding=1)
        self.conv5 = Conv2d(in_channels=384, out_channels=256, kernel_size=3, stride=1, padding=1)
        self.max_pool5 = MaxPool2d(kernel_size=2, stride=2)

        self.fc1 = Linear(in_features=12544, out_features=4096)
        self.drop_ratio1 = 0.5
        self.drop1 = Dropout(self.drop_ratio1)
        self.fc2 = Linear(in_features=4096, out_features=4096)
        self.drop_ratio2 = 0.5
        self.drop2 = Dropout(self.drop_ratio2)
        self.fc3 = Linear(in_features=4096, out_features=num_classes)
    
    def forward(self, x):
        x = self.conv1(x)
        x = F.relu(x)
        x = self.max_pool1(x)
        x = self.conv2(x)
        x = F.relu(x)
        x = self.max_pool2(x)
        x = self.conv3(x)
        x = F.relu(x)
        x = self.conv4(x)
        x = F.relu(x)
        x = self.conv5(x)
        x = F.relu(x)
        x = self.max_pool5(x)
        x = torch.reshape(x, [x.shape[0], -1])
        x = self.fc1(x)
        x = F.relu(x)
        # 在全连接之后使用dropout抑制过拟合
        x = self.drop1(x)
        x = self.fc2(x)
        x = F.relu(x)
        # 在全连接之后使用dropout抑制过拟合
        x = self.drop2(x)
        x = self.fc3(x)
        return x

# 创建模型
model = AlexNet(num_classes=1)
# 启动训练过程
opt = optim.Adam(lr=0.001, params=model.parameters())
train_pm(model, optimizer=opt)
evaluation(model, params_file_path="AlexNet.pdparams")

start training ... 
epoch: 0, batch_id: 0, loss is: 0.6922525763511658
epoch: 0, batch_id: 10, loss is: 0.7914059162139893
epoch: 0, batch_id: 20, loss is: 0.7226778864860535
epoch: 0, batch_id: 30, loss is: 0.5134126543998718
epoch: 0, batch_id: 40, loss is: 0.6680147647857666
epoch: 0, batch_id: 50, loss is: 0.09653188288211823
epoch: 0, batch_id: 60, loss is: 2.5109710693359375
[validation] accuracy/loss: 0.5649999976158142/0.911982536315918
epoch: 1, batch_id: 0, loss is: 1.3442769050598145
epoch: 1, batch_id: 10, loss is: 0.6957622766494751
epoch: 1, batch_id: 20, loss is: 0.19908615946769714
epoch: 1, batch_id: 30, loss is: 0.6529017686843872
epoch: 1, batch_id: 40, loss is: 0.4269561767578125
epoch: 1, batch_id: 50, loss is: 0.1539572924375534
epoch: 1, batch_id: 60, loss is: 0.38534554839134216
[validation] accuracy/loss: 0.9024999737739563/0.3134824335575104
epoch: 2, batch_id: 0, loss is: 0.1563398689031601
epoch: 2, batch_id: 10, loss is: 0.6282495260238647
epoch: 2, batch_id: 20, loss is: 0.3328627049922943
epoch: 2, batch_id: 30, loss is: 0.12043571472167969
epoch: 2, batch_id: 40, loss is: 0.33190542459487915
epoch: 2, batch_id: 50, loss is: 0.16366970539093018
epoch: 2, batch_id: 60, loss is: 0.3514131009578705
[validation] accuracy/loss: 0.8675000071525574/0.3479907512664795
epoch: 3, batch_id: 0, loss is: 0.47956809401512146
epoch: 3, batch_id: 10, loss is: 0.07707101106643677
epoch: 3, batch_id: 20, loss is: 0.21194897592067719
epoch: 3, batch_id: 30, loss is: 0.11190598458051682
epoch: 3, batch_id: 40, loss is: 0.004018225707113743
epoch: 3, batch_id: 50, loss is: 0.5130555033683777
epoch: 3, batch_id: 60, loss is: 0.38815149664878845
[validation] accuracy/loss: 0.9200000762939453/0.28641197085380554
epoch: 4, batch_id: 0, loss is: 0.40335366129875183
epoch: 4, batch_id: 10, loss is: 0.3855357766151428
epoch: 4, batch_id: 20, loss is: 0.16769567131996155
epoch: 4, batch_id: 30, loss is: 0.6962148547172546
epoch: 4, batch_id: 40, loss is: 1.2973268032073975
epoch: 4, batch_id: 50, loss is: 0.15464788675308228
epoch: 4, batch_id: 60, loss is: 0.012425403110682964
[validation] accuracy/loss: 0.8999999761581421/0.2859897017478943
start evaluation .......
loss=0.32200332209467886, acc=0.8800000078976155