esd文件转换成gho文件_CIFAR10数据集转换成图片及标注文件

一、CIFAR10的下载及使用

HUST小菜鸡：CIFAR10数据集下载​zhuanlan.zhihu.com

HUST小菜鸡：CIFAR10数据集的下载及使用​zhuanlan.zhihu.com

为了方便使用，我将转图片和转数据集标注的分开写，便于理解。后面我会附上同时转数据集为图片及标注文件的代码

二、将数据集转换成图片保存至指定文件夹

#官方给出的python3解压数据文件函数，返回数据字典

运行后将结果转换成了图片

D:Pycharmcifar10_classifiedvenvScriptspython.exe D:/Pycharm/cifar10_classified/trans2img.py

./datasets/cifar-10-batches-py/data_batch_1 is processing
./datasets/cifar-10-batches-py/data_batch_1 is done
./datasets/cifar-10-batches-py/data_batch_2 is processing
./datasets/cifar-10-batches-py/data_batch_2 is done
./datasets/cifar-10-batches-py/data_batch_3 is processing
./datasets/cifar-10-batches-py/data_batch_3 is done
./datasets/cifar-10-batches-py/data_batch_4 is processing
./datasets/cifar-10-batches-py/data_batch_4 is done
./datasets/cifar-10-batches-py/data_batch_5 is processing
./datasets/cifar-10-batches-py/data_batch_5 is done
./datasets/cifar-10-batches-py/test_batch is processing
./datasets/cifar-10-batches-py/test_batch is done
Finish transforming to image

三、将数据集标注生成文件保存

anno_loc = './datasets/annotations/'

#判断文件夹是否存在，不存在的话创建文件夹
if os.path.exists(anno_loc) == False:
    os.mkdir(anno_loc)

#用于存放图片文件名及标注
train_filenames = []
train_annotations = []

test_filenames = []
test_annotations= []

#训练集有五个批次，每个批次10000个图片，测试集有10000张图片
def cifar10_annotations(file_dir):
    print('creat train_img annotations')
    for i in range(1,6):
        data_name = file_dir + '/' + 'data_batch_' + str(i)
        data_dict = unpickle(data_name)
        print(data_name + ' is processing')
        for j in range(10000):
            img_name = str(data_dict[b'labels'][j]) + str((i) * 10000 + j) + '.jpg'
            img_annotations = data_dict[b'labels'][j]
            train_filenames.append(img_name)
            train_annotations.append(img_annotations)
        print(data_name + ' is done')

    test_data_name = file_dir + '/test_batch'
    print(test_data_name + ' is processing')
    test_dict = unpickle(test_data_name)

    for m in range(10000):
        testimg_name = str(test_dict[b'labels'][m]) + str(10000 + m) + '.jpg'
        testimg_annotations = test_dict[b'labels'][m]     #str(test_dict[b'labels'][m])    test_dict[b'labels'][m]
        test_filenames.append(testimg_name)
        test_annotations.append(testimg_annotations)

    print(test_data_name + ' is done')
    print('Finish file processing')


if __name__ == '__main__':

    file_dir = './datasets/cifar-10-batches-py'
    cifar10_annotations(file_dir)

    train_annot_dict = {
        'images': train_filenames,
        'categories': train_annotations
    }
    test_annot_dict = {
        'images':test_filenames,
        'categories':test_annotations
    }
    # print(annotation)

    train_json = json.dumps(train_annot_dict)
    train_file = open('./datasets/annotations/cifar10_train.json', 'w')
    train_file.write(train_json)
    train_file.close()

    test_json =json.dumps(test_annot_dict)
    test_file = open('./datasets/annotations/cifar10_test.json','w')
    test_file.write(test_json)
    test_file.close()
    print('annotations have writen to json file')

---

附转图片同时生成标注文件代码

import torch
import torchvision
import numpy as np
import cv2
import os
import json

cifar10 = torchvision.datasets.CIFAR10(
    root='datasets',
    train=True,
    download=False
)
cifar10_test = torchvision.datasets.CIFAR10(
    root='datasets',
    train=False,
    download=False
)
#输出数据集的信息
print(cifar10)
print(cifar10_test)

train_filenames = []
train_annotations = []
test_filenames = []
test_annotations= []

#cifar10 官方给出的解压函数
def unpickle(file):
    import pickle
    with open(file, 'rb') as fo:
        dict = pickle.load(fo, encoding='bytes')
    return dict

anno_loc = './datasets/annotations/'
loc_1 = './datasets/train_cifar10/'
loc_2 = './datasets/test_cifar10/'

#判断文件夹是否存在，不存在的话创建文件夹
if os.path.exists(loc_1) == False:
    os.mkdir(loc_1)
if os.path.exists(loc_2) == False:
    os.mkdir(loc_2)
if os.path.exists(anno_loc) == False:
    os.mkdir(anno_loc)


'''
def cifar10_annotations(file_dir):
    print('creat train_img annotations')
    for i in range(1,6):
        data_name = file_dir + '/' + 'data_batch_' + str(i)
        data_dict = unpickle(data_name)
        print(data_name + ' is processing')
        for j in range(10000):
            img_name = str(data_dict[b'labels'][j]) + str((i) * 10000 + j) + '.jpg'
            img_annotations = data_dict[b'labels'][j]
            train_filenames.append(img_name)
            train_annotations.append(img_annotations)
        print(data_name + ' is done')

    test_data_name = file_dir + '/test_batch'
    print(test_data_name + ' is processing')
    test_dict = unpickle(test_data_name)

    for m in range(10000):
        testimg_name = str(test_dict[b'labels'][m]) + str(10000 + m) + '.jpg'
        testimg_annotations = test_dict[b'labels'][m]     #str(test_dict[b'labels'][m])    test_dict[b'labels'][m]
        test_filenames.append(testimg_name)
        test_annotations.append(testimg_annotations)

    print(test_data_name + ' is done')
    print('Finish file processing')
'''

def cifar10_img(file_dir):
    for i in range(1,6):
        data_name = file_dir + '/'+'data_batch_'+ str(i)
        data_dict = unpickle(data_name)
        print(data_name + ' is processing')

        for j in range(10000):
            img = np.reshape(data_dict[b'data'][j],(3,32,32))
            img = np.transpose(img,(1,2,0))
            #通道顺序为RGB
            img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
            #要改成不同的形式的文件只需要将文件后缀修改即可
            img_name = loc_1 + str(data_dict[b'labels'][j]) + str((i)*10000 + j) + '.jpg'
            annot_img_name = str(data_dict[b'labels'][j]) + str((i)*10000 + j) + '.jpg'
            img_annotations = data_dict[b'labels'][j]
            train_filenames.append(annot_img_name)
            train_annotations.append(img_annotations)

            cv2.imwrite(img_name,img)

        print(data_name + ' is done')


    test_data_name = file_dir + '/test_batch'
    print(test_data_name + ' is processing')
    test_dict = unpickle(test_data_name)

    for m in range(10000):
        img = np.reshape(test_dict[b'data'][m], (3, 32, 32))
        img = np.transpose(img, (1, 2, 0))
        # 通道顺序为RGB
        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        # 要改成不同的形式的文件只需要将文件后缀修改即可
        img_name = loc_2 + str(test_dict[b'labels'][m]) + str(10000 + m) + '.jpg'
        annot_img_name = str(test_dict[b'labels'][m]) + str(10000 + m) + '.jpg'
        img_annotations = test_dict[b'labels'][m]
        test_filenames.append(annot_img_name)
        test_annotations.append(img_annotations)
        cv2.imwrite(img_name, img)
    print(test_data_name + ' is done')
    print('Finish transforming to image')

if __name__ == '__main__':

    file_dir = './datasets/cifar-10-batches-py'
    cifar10_img(file_dir)
    # cifar10_annotations(file_dir)

    train_annot_dict = {
        'images': train_filenames,
        'categories': train_annotations
    }
    test_annot_dict = {
        'images':test_filenames,
        'categories':test_annotations
    }
    # print(annotation)

    train_json = json.dumps(train_annot_dict)
    train_file = open('./datasets/annotations/cifar10_train.json', 'w')
    train_file.write(train_json)
    train_file.close()

    test_json =json.dumps(test_annot_dict)
    test_file = open('./datasets/annotations/cifar10_test.json','w')
    test_file.write(test_json)
    test_file.close()
    print('annotations have writen to json file')