pytorch加载自己的图片数据集的2种方法详解

导入ImageFolder()包
from torchvision.datasets import ImageFolder

import torch
from torchvision import transforms, datasets
import torch.nn as nn
from torch.utils.data import DataLoader
 
transforms = transforms.Compose([
    transforms.Resize(256),    # 将图片短边缩放至256，长宽比保持不变：
    transforms.CenterCrop(224),   #将图片从中心切剪成3*224*224大小的图片
    transforms.ToTensor()          #把图片进行归一化，并把数据转换成Tensor类型
]) 

path = r'D:\数据集\Flower_Orig_dataset'
data_train = datasets.ImageFolder(path, transform=transforms)
data_loader = DataLoader(data_train, batch_size=64, shuffle=True)
for i, data in enumerate(data_loader):
    images, labels = data
    print(images.shape)
    print(labels.shape)
    break

def make_txt(root, file_name, label):
    path = os.path.join(root, file_name)
    data = os.listdir(path)
    f = open(path+'\\'+'f.txt', 'w')
    for line in data:
        f.write(line+' '+str(label)+'\n')
    f.close()
#调用函数生成两个文件夹下的txt文件
make_txt(path, file_name='flower_orig', label=0)
make_txt(path, file_name='sunflower', label=1)

def link_txt(file1, file2):
    txt_list = []
    path = r'D:\数据集\Flower_Orig_dataset\data.txt'
 
    f = open(path, 'a')
 
    f1 = open(file1, 'r')
    data1 = f1.readlines()
    for line in data1:
        txt_list.append(line)
 
    f2 = open(file2, 'r')
    data2 = f2.readlines()
    for line in data2:
        txt_list.append(line)
 
    for line in txt_list:
        f.write(line)
 
    f.close()
    f1.close()
    f2.close()
 
#调用函数， 将两个文件夹下的txt文件合并
file1 = r'D:\数据集\Flower_Orig_dataset\flower_orig\f.txt'
file2 = r'D:\数据集\Flower_Orig_dataset\sunflower\f.txt'
link_txt(file1=file1, file2=file2)

我们读取图片的根目录， 在根目录下有所有图片的txt文件， 拿到txt文件后， 先读取txt文件， 之后遍历txt文件中的每一行， 首先去除掉尾部的换行符， 在以空格切分，前半部分是图片名称， 后半部分是图片标签， 当图片名称和根目录结合，就得到了我们的图片路径   
class MyDataset(Dataset):
    def __init__(self, img_path, transform=None):
        super(MyDataset, self).__init__()
        self.root = img_path
 
        self.txt_root = self.root + 'data.txt'
        f = open(self.txt_root, 'r')
        data = f.readlines()
 
        imgs = []
        labels = []
        for line in data:
            line = line.rstrip()
            Word = line.split()
            imgs.append(os.path.join(self.root, word[1], word[0]))
 
            labels.append(word[1])
        self.img = imgs
        self.label = labels
        self.transform = transform
 
    def __len__(self):
        return len(self.label)
 
    def __getitem__(self, item):
        img = self.img[item]
        label = self.label[item]
 
        img = Image.open(img).convert('RGB')
 
        #此时img是PIL.Image类型   label是str类型
 
        if transforms is not None:
            img = self.transform(img)
 
        label = np.array(label).astype(np.int64)
        label = torch.from_numpy(label)
        
        return img, label

path = r'D:\数据集\Flower_Orig_dataset'
dataset = MyDataset(path, transform=transform)
 
data_loader = DataLoader(dataset=dataset, batch_size=64, shuffle=True)

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(3,16,3)
        self.maxpool = nn.MaxPool2d(2,2)
        self.conv2 = nn.Conv2d(16,5,3)
 
        self.relu = nn.ReLU()
        self.fc1 = nn.Linear(55*55*5, 1200)
        self.fc2 = nn.Linear(1200,64)
        self.fc3 = nn.Linear(64,2)
 
    def forward(self,x):
        x = self.maxpool(self.relu(self.conv1(x)))    #113
        x = self.maxpool(self.relu(self.conv2(x)))    #55
        x = x.view(-1, self.num_flat_features(x))
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = self.fc3(x)
        return x
    
    
    def num_flat_features(self, x):
        size = x.size()[1:]
        num_features = 1
        for s in size:
            num_features *= s
 
        return num_features
 

model = Net()
 
criterion = torch.nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.01)
 
 
epochs = 10
for epoch in range(epochs):
    running_loss = 0.0
    for i, data in enumerate(data_loader):
        images, label = data
 
        out = model(images)
 
        loss = criterion(out, label)
 
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
 
        running_loss += loss.item()
        if(i+1)%10 == 0:
            print('[%d  %5d]   loss: %.3f'%(epoch+1, i+1, running_loss/100))
            running_loss = 0.0
 
print('finished train')

#保存模型
torch.save(net, 'model_name.pth')   #保存的是模型， 不止是w和b权重值
 
# 读取模型
model = torch.load('model_name.pth')