使用python/pytorch读取数据集的示例代码

import torch
from torch.utils.data import Dataset
from torchvision.datasets import MNIST
from torch.utils.data import DataLoader
from tqdm import tqdm
from torchvision import models, transforms
from torchvision.utils import save_image
from PIL import Image

#将MNIST数据集转换为图片
tf = transforms.Compose([transforms.ToTensor()]) # mnist is already normalised 0 to 1
datasetMNIST = MNIST("./data", train=True, download=True, transform=tf)
pbar = tqdm(datasetMNIST)
for index, (img,cl) in enumerate(pbar):
   save_image(img, f"./data/MNIST_PNG/x/{index}.png")
   # 以写入模式打开文件
   with open(f"./data/MNIST_PNG/c/{index}.txt", "w", encoding="utf-8") as file:
        # 将字符串写入文件
        file.write(f"{cl}")

class MyMNISTDataset(Dataset):
   def __init__(self, data):
       self.data = data

   def __len__(self):
       return len(self.data)

   def __getitem__(self, idx):
       x = self.data[idx][0] #图像
       y = self.data[idx][1] #标签
       return x, y
   
def load_data(dataNum=60000):
    data = []
    pbar = tqdm(range(dataNum))
    for i in pbar:
        # 指定图片路径
        image_path = f'./data/MNIST_PNG/x/{i}.png'
        cond_path=f'./data/MNIST_PNG/c/{i}.txt'
        # 定义图像预处理
        preprocess = transforms.Compose([
        transforms.Grayscale(num_output_channels=1),  # 将图像转换为灰度图像（单通道）
        transforms.ToTensor()
        ])
        # 使用预处理加载图像
        image_tensor = preprocess(Image.open(image_path))
        # 加载条件文档（tag）
        with open(cond_path, 'r') as file:
            line = file.readline()
            number = int(line)  # 将字符串转换为整数，图像的类别
            data.append((image_tensor, number))
    return data
   

data=load_data(60000)
# 创建数据集实例
dataset = MyMNISTDataset(data)

# 创建数据加载器
dataloader = DataLoader(dataset, batch_size=4, shuffle=True)
pbar = tqdm(dataloader)

for index, (img,cond) in enumerate(pbar):
    #这里对每一批进行训练...
    print(f"Batch {index}: img = {img.shape}, cond = {cond}")

......
Batch 7847: img = torch.Size([4, 1, 28, 28]), cond = tensor([0, 1, 5, 2])
Batch 7848: img = torch.Size([4, 1, 28, 28]), cond = tensor([2, 2, 6, 0])
Batch 7849: img = torch.Size([4, 1, 28, 28]), cond = tensor([4, 3, 0, 9])
Batch 7850: img = torch.Size([4, 1, 28, 28]), cond = tensor([6, 2, 9, 5])
Batch 7851: img = torch.Size([4, 1, 28, 28]), cond = tensor([7, 2, 4, 4])
Batch 7852: img = torch.Size([4, 1, 28, 28]), cond = tensor([1, 4, 2, 6])
Batch 7853: img = torch.Size([4, 1, 28, 28]), cond = tensor([2, 5, 3, 5])
Batch 7854: img = torch.Size([4, 1, 28, 28]), cond = tensor([7, 1, 0, 1])
Batch 7855: img = torch.Size([4, 1, 28, 28]), cond = tensor([9, 8, 9, 7])
Batch 7856: img = torch.Size([4, 1, 28, 28]), cond = tensor([4, 6, 6, 7])
Batch 7857: img = torch.Size([4, 1, 28, 28]), cond = tensor([7, 4, 1, 6])
Batch 7858: img = torch.Size([4, 1, 28, 28]), cond = tensor([5, 4, 6, 5])
Batch 7859: img = torch.Size([4, 1, 28, 28]), cond = tensor([6, 3, 1, 9])
Batch 7860: img = torch.Size([4, 1, 28, 28]), cond = tensor([5, 5, 8, 6])
Batch 7861: img = torch.Size([4, 1, 28, 28]), cond = tensor([0, 4, 8, 9])
Batch 7862: img = torch.Size([4, 1, 28, 28]), cond = tensor([2, 3, 5, 8])
Batch 7863: img = torch.Size([4, 1, 28, 28]), cond = tensor([8, 0, 0, 6])
......