【无标题】

准备

构架一个目录结构即可

在MNIST_IMG文件夹中建两个文件夹分别是TEST和TRAIN，然后每个文件夹里面建名字为0到9 的10个文件夹即可


TRAIN文件夹类似就步粘贴了

代码

测试集

import torch
import torch.nn as nn
import numpy as np
from torchvision import datasets
from torchvision import transforms
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt
import cv2

batch_size = 1
transform = transforms.Compose([
    transforms.ToTensor(),
])

train_dataset = datasets.MNIST(
    root='./MNIST_IMG/mnist/',
    train=True,
    download=True,
    transform=transform
)
train_loader = DataLoader(
    dataset=train_dataset,
    shuffle=True,
    batch_size=batch_size
)
test_dataset = datasets.MNIST(
    root='./MNIST_IMG/mnist/',
    train=False,
    download=True,
    transform=transform
)
test_loader = DataLoader(
    dataset=test_dataset,
    shuffle=False,
    batch_size=batch_size
)

def save_train_data():
    num0 = 0
    num1 = 0
    num2 = 0
    num3 = 0
    num4 = 0
    num5 = 0
    num6 = 0
    num7 = 0
    num8 = 0
    num9 = 0

    for batch_idx, data in enumerate(test_loader, 0):
        inputs, targets = data
        inputs = inputs.numpy()
        targets = targets.numpy()

        inputs = inputs.reshape(28, 28)*256
        if targets == 0:
            cv2.imwrite('./MNIST_IMG/TEST/0/' + str(num0) + '.jpg',inputs)
            num0=num0+1

        if targets == 1:
            cv2.imwrite('./MNIST_IMG/TEST/1/' + str(num1) + '.jpg',inputs)
            num1=num1+1

        if targets == 2:
            cv2.imwrite('./MNIST_IMG/TEST/2/' + str(num2) + '.jpg',inputs)
            num2=num2+1

        if targets == 3:
            cv2.imwrite('./MNIST_IMG/TEST/3/' + str(num3) + '.jpg',inputs)
            num3=num3+1

        if targets == 4:
            cv2.imwrite('./MNIST_IMG/TEST/4/' + str(num4) + '.jpg',inputs)
            num4=num4+1

        if targets == 5:
            cv2.imwrite('./MNIST_IMG/TEST/5/' + str(num5) + '.jpg',inputs)
            num5=num5+1

        if targets == 6:
            cv2.imwrite('./MNIST_IMG/TEST/6/' + str(num6) + '.jpg',inputs)
            num6=num6+1

        if targets == 7:
            cv2.imwrite('./MNIST_IMG/TEST/7/' + str(num7) + '.jpg',inputs)
            num7=num7+1

        if targets == 8:
            cv2.imwrite('./MNIST_IMG/TEST/8/' + str(num8) + '.jpg',inputs)
            num8=num8+1

        if targets == 9:
            cv2.imwrite('./MNIST_IMG/TEST/9/' + str(num9) + '.jpg', inputs)
            num9=num9+1

        print(str(batch_idx) + '.jpg' + '-' + str(targets))

if __name__ == '__main__':
    save_train_data()

训练集

import torch
import torch.nn as nn
import numpy as np
from torchvision import datasets
from torchvision import transforms
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt
import cv2

batch_size = 1
transform = transforms.Compose([
    transforms.ToTensor()
])

train_dataset = datasets.MNIST(
    root='./MNIST_IMG/mnist/',
    train=True,
    download=True,
    transform=transform
)
train_loader = DataLoader(
    dataset=train_dataset,
    shuffle=True,
    batch_size=batch_size
)
test_dataset = datasets.MNIST(
    root='./MNIST_IMG/mnist/',
    train=False,
    download=True,
    transform=transform
)
test_loader = DataLoader(
    dataset=test_dataset,
    shuffle=False,
    batch_size=batch_size
)

def save_train_data():
    num0 = 0
    num1 = 0
    num2 = 0
    num3 = 0
    num4 = 0
    num5 = 0
    num6 = 0
    num7 = 0
    num8 = 0
    num9 = 0

    for batch_idx, data in enumerate(train_loader, 0):
        inputs, targets = data
        inputs = inputs.numpy()
        targets = targets.numpy()

        inputs = inputs.reshape(28, 28)*256
        if targets == 0:
            cv2.imwrite('./MNIST_IMG/TRAIN/0/' + str(num0) + '.jpg',inputs)
            num0=num0+1

        if targets == 1:
            cv2.imwrite('./MNIST_IMG/TRAIN/1/' + str(num1) + '.jpg',inputs)
            num1=num1+1

        if targets == 2:
            cv2.imwrite('./MNIST_IMG/TRAIN/2/' + str(num2) + '.jpg',inputs)
            num2=num2+1

        if targets == 3:
            cv2.imwrite('./MNIST_IMG/TRAIN/3/' + str(num3) + '.jpg',inputs)
            num3=num3+1

        if targets == 4:
            cv2.imwrite('./MNIST_IMG/TRAIN/4/' + str(num4) + '.jpg',inputs)
            num4=num4+1

        if targets == 5:
            cv2.imwrite('./MNIST_IMG/TRAIN/5/' + str(num5) + '.jpg',inputs)
            num5=num5+1

        if targets == 6:
            cv2.imwrite('./MNIST_IMG/TRAIN/6/' + str(num6) + '.jpg',inputs)
            num6=num6+1

        if targets == 7:
            cv2.imwrite('./MNIST_IMG/TRAIN/7/' + str(num7) + '.jpg',inputs)
            num7=num7+1

        if targets == 8:
            cv2.imwrite('./MNIST_IMG/TRAIN/8/' + str(num8) + '.jpg',inputs)
            num8=num8+1

        if targets == 9:
            cv2.imwrite('./MNIST_IMG/TRAIN/9/' + str(num9) + '.jpg', inputs)
            num9=num9+1

        print(str(batch_idx) + '.jpg' + '-' + str(targets))

if __name__ == '__main__':
    save_train_data()

效果就不说了，没意思