解析MNIST的数据集

MNIST的官网数据集一直下不下来，好像要翻墙，总之查找不到URL，所以自己找其他途径下了，结果下下来不是整理好的数据集，长这样

后来才知道可以用struct函数来解析这种格式的文件，解析出来一堆jpg的图片，但是又不会用了，总之先把struct的代码放上来吧，省的以后忘了找不着

'''
from PIL import Image
import struct

def read_image(filename):
  f = open(filename, 'rb')
  index = 0
  buf = f.read()
  f.close()
  magic, images, rows, columns = struct.unpack_from('>IIII' , buf , index)
  index += struct.calcsize('>IIII')
  for i in range(10000):
  #for i in xrange(2000):
    image = Image.new('L', (columns, rows))

    for x in range(rows):
      for y in range(columns):
        image.putpixel((y, x), int(struct.unpack_from('>B', buf, index)[0]))
        index += struct.calcsize('>B')

    print ('save ' + str(i) + 'image')
    image.save('C:/Users/cxt66/Desktop/MNISTdata/MNISTdata/test_images/test' + str(i) + '.png')

def read_label(filename, saveFilename):
  f = open(filename, 'rb')
  index = 0
  buf = f.read()

  f.close()

  magic, labels = struct.unpack_from('>II' , buf , index)
  index += struct.calcsize('>II')
  
  labelArr = [0] * labels
  #labelArr = [0] * 2000

  for x in range(60000):
  #for x in xrange(2000):
    labelArr[x] = int(struct.unpack_from('>B', buf, index)[0])
    index += struct.calcsize('>B')

  save = open(saveFilename, 'w')

  save.write(','.join(map(lambda x: str(x), labelArr)))
  save.write('\n')

  save.close()
  print ('save labels success')

if __name__ == '__main__':
  read_image('C:/Users/cxt66/Desktop/MNISTdata/MNISTdata/t10k-images.idx3-ubyte')
  read_label('C:/Users/cxt66/Desktop/MNISTdata/MNISTdata/train-labels.idx1-ubyte','C:/Users/cxt66/Desktop/MNISTdata/MNISTdata/train_label.txt')

 

'''