cifar10图片可视化(Win7+python3.5)

1. 将%caffe_root%\data\cifar10\get_cifar10.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batches.meta.txt，data_batch_1.bin，data_batch_2.bin，data_batch_3.bin，data_batch_4.bin，data_batch_5.bin，test_batch.bin。batches为标签对应的类别名；data_batch_x.bin为第x批10000张的训练图片，共5批；test_batch.bin为10000张测试图片。
2. 创建%caffe_root%\data\cifar10\cifar10_images\test\data_batch_x（x为1-5共5个文件夹）及%caffe_root%\data\mnist\mnist_images\test_batch文件夹，以保存转换后的图片
3. 编写如下python脚本，并运行。在对应文件夹下得到图片文件。

cifar10_data_root = r'd:/caffe/data/cifar10/'

import struct
import numpy as np
from PIL import Image

def load_CIFAR_Labels(filename):
    with open(filename, 'r') as f:
        lines = []
        for line in f.readlines():
            lines.append(line.strip('\n'))
        return lines

def load_CIFAR_batch(filename, label, Pic):
    index = 0
    bin_file = open(filename, 'rb')
    buf = bin_file.read()
    for i in range(10000):
        (l,) = struct.unpack_from('>b', buf, index)
        label.append(l)
        index += struct.calcsize('>b')
        Pic.append(struct.unpack_from('>3072B', buf, index))
        index += struct.calcsize('>3072B')

if __name__ == "__main__":
    lines = load_CIFAR_Labels(cifar10_data_root + "batches.meta.txt")
    print(lines)
    NUM_DATA = 10000

    ##### TEST DATA SHOW #####
    dataPath = cifar10_data_root + r"test_batch.bin"
    print("### " + dataPath + " is loading... ###")

    label = []
    Pic = []
    count = [0] * 10
    load_CIFAR_batch(dataPath, label, Pic)

    for i in range(NUM_DATA):
        im = np.array(Pic[i]).reshape(3, 32, 32)
        image = Image.new("RGB", (32, 32))
        for channel in range(3):
            for a in range(32):
                for b in range(32):
                    image.putpixel((a, b), (im[0][b][a], im[1][b][a], im[2][b][a]))

        picName = cifar10_data_root + "cifar10_images/test_batch/test_" + str(i) + "_" + lines[label[i]] + "_" + str(count[label[i]]) + ".png"
        count[label[i]] += 1
        image.save(picName, "png")

        if i % 500 == 0:
            print("testSet processed:%d" % i)

    ##### TRAIN DATA #####
    for j in range(1, 6):
        dataPath = cifar10_data_root + r"data_batch_%d.bin"%1
        print("### " + dataPath + " is loading... ###")

        label = []
        Pic = []
        count = [0] * 10
        load_CIFAR_batch(dataPath, label, Pic)

        for i in range(NUM_DATA):
            im = np.array(Pic[i]).reshape(3, 32, 32)
            image = Image.new("RGB", (32, 32))
            for channel in range(3):
                for a in range(32):
                    for b in range(32):
                        image.putpixel((a, b), (im[0][b][a], im[1][b][a], im[2][b][a]))

            picName = cifar10_data_root + "cifar10_images/data_batch_%d/train[%d]_"%(j,j) + str(i) + "_" + lines[label[i]] + "_" + str(count[label[i]]) + ".png"
            count[label[i]] += 1
            image.save(picName, "png")

            if i % 500 == 0:
                print("trainSet processed:patch(%d) %d" % (j,i))

最后生成文件总占用空间234M。