MNIST图片可视化(Win7+python3.5)

最新推荐文章于 2021-03-25 20:02:28 发布

VOY_KANG

最新推荐文章于 2021-03-25 20:02:28 发布

阅读量702

点赞数

本文链接：https://blog.youkuaiyun.com/rono_back/article/details/77799144

版权

python 同时被 2 个专栏收录

8 篇文章

订阅专栏

caffe

7 篇文章

订阅专栏

在Win7环境下，使用Python3.5和Caffe，通过转换Linux格式的get_mnist.sh脚本，利用Cygwin下载并解压MNIST数据集。将原始数据转存为图片文件，分别保存在test和train文件夹中，最终生成的图片文件总大小为273M。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

将%caffe_root%\data\mnist\get_mnist.sh转换成linux格式，并用cygwin运行（事先装好wget）。自动下载并解压，得到t10k-images-idx3-ubyte，t10k-labels-idx1-ubyte，train-images-idx3-ubyte，train-labels-idx1-ubyte
创建%caffe_root%\data\mnist\mnist_images\test及%caffe_root%\data\mnist\mnist_images\train文件夹，以保存转换后的图片
编写如下python脚本，并运行。在对应文件夹下得到图片文件。

import numpy as np
import struct
from PIL import Image

caffe_root = r"D:/caffe"
caffe_mnist_data = caffe_root + r"/data/mnist/"

test_MNIST_data = caffe_mnist_data + r"t10k-images-idx3-ubyte"
train_MNIST_data = caffe_mnist_data + r"train-images-idx3-ubyte"
test_MNIST_label = caffe_mnist_data + r"t10k-labels-idx1-ubyte"
train_MNIST_label = caffe_mnist_data + r"train-labels-idx1-ubyte"

### LABEL ###
index = 0
print("TEST LABEL DATA:" + test_MNIST_label)
binfile = open(test_MNIST_label, 'rb')
buf = binfile.read()
magic, numImages = struct.unpack_from('>II', buf, index)
print(magic, numImages)
index += struct.calcsize('>II')

label_test = []
for i in range(numImages):
    if i%1000 == 0:
        print("test label:" + str(i))
    (l,) = struct.unpack_from('>B', buf, index)
    index += struct.calcsize('>B')
    label_test.append(l)

index = 0
print("TRAIN LABEL DATA:" + train_MNIST_label)
binfile = open(train_MNIST_label, 'rb')
buf = binfile.read()
magic, numImages = struct.unpack_from('>II', buf, index)
print(magic, numImages)
index += struct.calcsize('>II')

label_train = []
for i in range(numImages):
    if i%1000 == 0:
        print("test label:" + str(i))
    (l,) = struct.unpack_from('>B', buf, index)
    index += struct.calcsize('>B')
    label_train.append(l)

### DATA ###
# TEST
index = 0
print("TEST IMAGE DATA:" + test_MNIST_data)
binfile = open(test_MNIST_data, 'rb')
buf = binfile.read()
magic, numImages, numRows, numColumns = struct.unpack_from('>IIII', buf, index) #4 * Int
print(magic, numImages, numRows, numColumns)
index += struct.calcsize('>IIII')

FileDir = caffe_mnist_data + r"mnist_images/test/"
print(FileDir)
for i in range(numImages):
    if i%1000 == 0:
        print("test data:" + str(i))

    im = struct.unpack_from('>784B', buf, index)    #28*28=784 Byte
    index += struct.calcsize('>784B')
    im = np.array(im)
    im = im.reshape(28, 28)

    picName = FileDir + r'\test_%d_label_%d.bmp'%(i,label_test[i])
    image = Image.new("L", (28,28))
    for a in range(28):
        for b in range(28):
            image.putpixel((a,b), int(im[b][a]))
    image.save(picName, 'bmp')


# TRAIN
index = 0
print("TRAIN IMAGE DATA:" + train_MNIST_data)
binfile = open(train_MNIST_data, 'rb')
buf = binfile.read()
magic, numImages, numRows, numColumns = struct.unpack_from('>IIII', buf, index)
print(magic, numImages, numRows, numColumns)
index += struct.calcsize('>IIII')

FileDir = caffe_mnist_data + r"mnist_images/train/"
print(FileDir)
ls = os.listdir(FileDir)
if len(ls) >= numImages:
    numImages = 0
for i in range(numImages):
    if i%1000 == 0:
        print("train:" + str(i))

    im = struct.unpack_from('>784B', buf, index)
    index += struct.calcsize('>784B')
    im = np.array(im)
    im = im.reshape(28, 28)

    picName = FileDir + r'\train_%d_label_%d.bmp'%(i, label_train[i])
    image = Image.new("L", (28,28))
    for a in range(28):
        for b in range(28):
            image.putpixel((a,b), int(im[b][a]))
    image.save(picName, 'bmp')

最后生成文件总占用空间273M。