[Paddle学习笔记][02][MNIST转换到PNG]

说明：

MNIST手写数字数据集非常经典，它由60000个训练样本和10000个测试样本组成，每个样本都是一张28×28的灰度图片。直接下载下来的MNIST数据是无法通过解压或应用程序打开的，因为它里面的文件不是标准的图像格式，而是以字节的形式存储的，所以需要借助编程的手段来打开。

实验代码：

 

import os
import struct
import png
from array import array

# 文件路径
train_images = './train-images-idx3-ubyte'
train_labels = './train-labels-idx1-ubyte' 
test_images = './t10k-images-idx3-ubyte'
test_labels = './t10k-labels-idx1-ubyte'
train_folder = './train'
test_folder = './test'

# 创建目录
if not os.path.exists(train_folder):
    os.makedirs(train_folder)
if not os.path.exists(test_folder):
    os.makedirs(test_folder)
    
train_folders = [os.path.join(train_folder, str(i)) for i in range(10)]
test_folders = [os.path.join(test_folder, str(i)) for i in range(10)]

for dir in train_folders:
    if not os.path.exists(dir):
        os.makedirs(dir)
for dir in test_folders:
    if not os.path.exists(dir):
        os.makedirs(dir)
    
# 打开文件
train_imgs = open(train_images, 'rb') # 以二进制读模式打开文件
train_labs = open(train_labels, 'rb')
test_imgs = open(test_images, 'rb')
test_labs = open(test_labels, 'rb')

# 读取数据
struct.unpack('>IIII', train_imgs.read(16)) # 大端模式，每次读取16字节
struct.unpack('>II', train_labs.read(8)) # 大端模式，每次读取8字节
struct.unpack('>IIII', test_imgs.read(16))
struct.unpack('>II', test_labs.read(8))

train_img = array('B', train_imgs.read()) # 无符号字节array数组类型
train_lab = array('b', train_labs.read()) # 有符号字节array数据类型
test_img = array('B', test_imgs.read())
test_lab = array('b', test_labs.read())

train_imgs.close() # 关闭文件
train_labs.close()
test_imgs.close()
test_imgs.close()

# 保存图像
for (i, label) in enumerate(train_lab):
    filename = os.path.join(train_folders[label], str(i) + '.png')
    print('writing ' + filename)
    with open(filename, 'wb') as img:
        image = png.Writer(28, 28, greyscale=True)
        data = [train_img[(i*28*28 + j*28) : (i*28*28 + (j+1)*28)] for j in range(28)]
        image.write(img, data) # 保存训练图像
for (i, label) in enumerate(test_lab):
    filename = os.path.join(test_folders[label], str(i) + '.png')
    print('writing ' + filename)
    with open(filename, 'wb') as img:
        image = png.Writer(28, 28, greyscale=True)
        data = [test_img[(i*28*28 + j*28) : (i*28*28 + (j+1)*28)] for j in range(28)]
        image.write(img, data) # 保存测试图像

 

实验结果：

参考资料：

https://blog.csdn.net/SongGu1996/article/details/98849274