原始代码

import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision
from torchvision import datasets, transforms
import os, PIL, pathlib,random


device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

data_dir = '/Users/montylee/NJUPT/Learn/Github/deeplearning/pytorch/P6/data/'
data_dir = pathlib.Path(data_dir)

data_paths = list(data_dir.glob('*'))
classNames = [str(path).split('/')[-1] for path in data_paths]

train_transforms = transforms.Compose([
    transforms.Resize([224,224]),
    transforms.RandomHorizontalFlip(), # 随机翻转和旋转
    transforms.ToTensor(),
    transforms.Normalize(mean = [0.485, 0.456, 0.406], # 均值
                         std = [0.229, 0.224, 0.225]) # 方差
])

total_dataset = datasets.ImageFolder(data_dir, transform=train_transforms)

total_dataset.class_to_idx

train_size = int(0.8 * len(total_dataset))
test_size = len(total_dataset) - train_size
train_dataset, test_dataset = torch.utils.data.random_split(total_dataset, [train_size, test_size])
train_dataset, test_dataset

batch_size = 32
train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=1)
test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=batch_size, shuffle=True, num_workers=1)
[32]
# 构建 CNN 网络
import torch.nn.functional as F

num_classes = len(classNames)

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        # 特征提取网络
        self.conv1 = nn.Conv2d(3,16,kernel_size=3)
        self.bn1 = nn.BatchNorm2d(16)
        self.pool1 = nn.MaxPool2d(kernel_size=2)
        self.dropout = nn.Dropout(p=0.3)

        self.conv2 = nn.Conv2d(16,32,kernel_size=3)
        self.bn2 = nn.BatchNorm2d(32)
        self.pool2 = nn.MaxPool2d(kernel_size=2)
        self.dropout = nn.Dropout(p=0.3)

        self.conv3 = nn.Conv2d(32,64,kernel_size=3)
        self.bn3 = nn.BatchNorm2d(64)
        self.pool3 = nn.MaxPool2d(kernel_size=2)
        # self.dropout = nn.Dropout(p=0.3)

        # self.conv4 = nn.Conv2d(128,256,kernel_size=3)
        # self.pool4 = nn.MaxPool2d(kernel_size=2)


        # 分类网络
        self.fc1 = nn.Linear(64*26*26, 256)
        self.fc2 = nn.Linear(256, num_classes)


    def forward(self, x):
        x = self.pool1(F.relu(self.conv1(x)))
        x = self.dropout(x)
        x = self.pool2(F.relu(self.conv2(x)))
        x = self.dropout(x)
        x = self.pool3(F.relu(self.conv3(x)))
        # x = self.dropout(x)
        # x = self.pool4(F.relu(self.conv4(x)))

        # print(x.shape)

        x = torch.flatten(x, start_dim=1)

        x = F.relu(self.fc1(x))
        x = self.fc2(x)

        return x


loss_fn = nn.CrossEntropyLoss()
learn_rate = 1e-2
opt = torch.optim.SGD(model.parameters(), lr=learn_rate)

def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)
    num_batches = len(dataloader)

    train_loss, train_acc = 0 , 0

    for x , y in dataloader:
        x, y = x.to(device), y.to(device)

        pred = model(x)
        loss = loss_fn(pred, y)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
        train_loss += loss.item()

    train_acc /= size
    train_loss /= num_batches

    return train_acc, train_loss


def test(dataloader, model, loss_fn):
    size = len(dataloader.dataset)
    num_batches = len(dataloader)

    test_loss, test_acc = 0 , 0

    with torch.no_grad():
        for imgs, target in dataloader:
            imgs, target = imgs.to(device), target.to(device)

            target_pred = model(imgs)
            loss = loss_fn(target_pred, target)

            test_loss += loss.item()
            test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item()

    test_acc /= size
    test_loss /= num_batches

    return test_acc, test_loss


epochs = 10
train_loss = []
train_acc = []
test_loss = []
test_acc = []

for epoch in range(epochs):
    model.train()
    epoch_train_acc, epoch_train_loss = train(train_loader, model, loss_fn, opt)

    model.eval()
    epoch_test_acc, epoch_test_loss = test(test_loader, model, loss_fn)

    train_acc.append(epoch_train_acc)
    train_loss.append(epoch_train_loss)
    test_acc.append(epoch_test_acc)
    test_loss.append(epoch_test_loss)

    template = ('Epoch:{:2d},Train_acc:{:.1f}%,Train_loss:{:.3f},Test_acc:{:.1f}%,Test_loss:{:.3f}')
    print(template.format(epoch+1,epoch_train_acc*100,epoch_train_loss,epoch_test_acc*100,epoch_test_loss))

print("Done")

import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')
plt.rcParams['font.sans-serif'] = ['SimHei']
plt.rcParams['axes.unicode_minus'] = False
plt.rcParams['figure.dpi'] = 100

epochs_range = range(epochs)

plt.figure(figsize = (12,3))
plt.subplot(1,2,1)
plt.plot(epochs_range,train_acc,label='Training Accuracy')
plt.plot(epochs_range,test_acc,label='Test Accuracy')
plt.legend()
plt.title("Training and Validation Accuracy")

plt.subplot(1,2,2)
plt.plot(epochs_range,train_loss,label="Training Loss")
plt.plot(epochs_range,test_loss,label='Testing Loss')
plt.legend()
plt.title('Training and Validation Loss')
plt.show()

结果：

问题：

1.test_accuracy太低，小于30%，且没有收敛

2.只训练了10个epoch

解决方法

数据增强

数据集数量太少，而且将数据集分为训练集与测试集之后，数量更为稀少。

可以通过平移、旋转、缩放、裁剪、翻转、变形等，对图片进行处理，增加数据集数量

代码

（下面只对图像进行了翻转）

def data_aug(input_path, output_path,  new_data_path):
    imgaug(input_path, output_path)
    rename_img(output_path)
    merge_files(new_data_path, input_path, output_path)

def imgaug(input_path, output_path):
    input_dirs = [d for d in os.listdir(input_path) if os.path.isdir(os.path.join(input_path, d))]
    output_dirs = [d for d in os.listdir(output_path) if os.path.isdir(os.path.join(output_path, d))]

    for i in range(len(input_dirs)):
        input_dirs[i] = input_path + "\\" + input_dirs[i]

    for i in range(len(output_dirs)):
        output_dirs[i] = output_path + "\\" + output_dirs[i]

    seq = iaa.Sequential([iaa.Flipud(1.0)])

    len_class = len(input_dirs)

    for i in range(len_class):
        for file_name in os.listdir(input_dirs[i]):

            file_path = os.path.join(input_dirs[i], file_name)
            img = Image.open(file_path)


            img_arr = np.array(img)
            img_aug = seq(images=img_arr)
            img_aug = Image.fromarray(img_aug)

            if img_aug.mode == 'RGBA':
                    img_aug = img_aug.convert('RGB')

            output_file_path = os.path.join(output_dirs[i], file_name)
            img_aug.save(output_file_path)

def rename_img(folder_path):

    counters = {} 

    for folder_name in os.listdir(folder_path):
        folder = os.path.join(folder_path, folder_name)
        if os.path.isdir(folder):
            file_types = ['*.jpg', '*.jpeg', '*.png'] 
            total_files = 0
            for file_type in file_types:
                total_files += len(glob.glob(os.path.join(folder, file_type)))
            counters[folder_name] = total_files 
    counters = {key: value + 1 for key, value in counters.items()}

    for subfolder in os.listdir(folder_path):
        subfolder_path = os.path.join(folder_path, subfolder)
        if os.path.isdir(subfolder_path):
            counter = counters[subfolder]

            for name in os.listdir(subfolder_path):
                if name.endswith('.jpg') or name.endswith('.jpeg') or name.endswith('.png')
                    keyword = subfolder
                    number = counter
                    extension = os.path.splitext(name)[1][1:]
                    new_name = keyword + str(number) + '.' + extension
                    os.rename(os.path.join(subfolder_path, name), os.path.join(subfolder_path, new_name)）
                    counter += 1


def merge_files(new_data_path, input_path, output_path):
    new_data_dirs = [d for d in os.listdir(new_data_path) if os.path.isdir(os.path.join(new_data_path, d))]

    for i in range(len(new_data_dirs)):
        new_data_dirs[i] = new_data_path + "\\" + new_data_dirs[i]

    len_class = len(new_data_dirs)
    input_dirs = [d for d in os.listdir(input_path) if os.path.isdir(os.path.join(input_path, d))]
    output_dirs = [d for d in os.listdir(output_path) if os.path.isdir(os.path.join(output_path, d))]

    for i in range(len(input_dirs)):
        input_dirs[i] = input_path + "\\" + input_dirs[i]

    for i in range(len(output_dirs)):
        output_dirs[i] = output_path + "\\" + output_dirs[i]
    for i in range(len_class):
       
        for filename in os.listdir(input_dirs[i]):
            if filename.endswith(".jpg"): 
                src_path = os.path.join(input_dirs[i], filename)
                dst_path = os.path.join(new_data_dirs[i], filename)
                shutil.copyfile(src_path, dst_path)

        for filename in os.listdir(output_dirs[i]):
            if filename.endswith(".jpg"):  
                src_path = os.path.join(output_dirs[i], filename)
                dst_path = os.path.join(new_data_dirs[i], filename)
                shutil.copyfile(src_path, dst_path)
if name == "__main__":

    input_path = r"C:\Users\STARRY\PycharmProjects\pythonProject1\data"# 原数据集路径
    output_path = r"C:\Users\STARRY\PycharmProjects\pythonProject1\data2" # 反转后存放数据集的路径
    new_data_path = r"C:\Users\STARRY\PycharmProjects\pythonProject1\data3" # 最终两个数据集合并后的路径

    data_aug(input_path, output_path,  new_data_path)

 结果

epoch=10

epoch=20

 建立更好的神经网络

1.可以使用ResNet50或者VGG16网络模型来替代

2.这里我是直接在原有的网络结构基础上加了

self.dropout = nn.Dropout(p=0.3)
self.conv4 = nn.Conv2d(128,256,kernel_size=3)
self.pool4 = nn.MaxPool2d(kernel_size=2)

代码

（只有神经网络部分做了改动）

class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(3,16,kernel_size=3)
        self.bn1 = nn.BatchNorm2d(16)
        self.pool1 = nn.MaxPool2d(kernel_size=2)
        self.dropout = nn.Dropout(p=0.3)

        self.conv2 = nn.Conv2d(16,32,kernel_size=3)
        self.bn2 = nn.BatchNorm2d(32)
        self.pool2 = nn.MaxPool2d(kernel_size=2)
        self.dropout = nn.Dropout(p=0.3)

        self.conv3 = nn.Conv2d(32,128,kernel_size=3)
        self.bn3 = nn.BatchNorm2d(64)
        self.pool3 = nn.MaxPool2d(kernel_size=2)
        self.dropout = nn.Dropout(p=0.3)

        self.conv4 = nn.Conv2d(128,256,kernel_size=3)
        self.pool4 = nn.MaxPool2d(kernel_size=2)
        self.fc1 = nn.Linear(36864, 256)
        self.fc2 = nn.Linear(256, num_classes)


    def forward(self, x):
        x = self.pool1(F.relu(self.conv1(x)))
        x = self.dropout(x)
        x = self.pool2(F.relu(self.conv2(x)))
        x = self.dropout(x)
        x = self.pool3(F.relu(self.conv3(x)))
        x = self.dropout(x)
        x = self.pool4(F.relu(self.conv4(x))
        x = torch.flatten(x, start_dim=1)
        x = F.relu(self.fc1(x))
        x = self.fc2(x)

        return x

结果

epoch=20

这里最高是到了85.4%

调整一些参数

可以调节优化器的学习速率以及batc_size的大小去提高准确度