CIFAR-10 数据转为图片-python

本文主要是介绍CIFAR-10 数据转为图片-python，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！

CIFAR-10 数据集

"""
CIFAR-10 是 32X32 的彩色图片，共有10个类别，每个类别6000张图片，50000张训练图片（均分为5个batch），10000张测试图片（每个类别选1000张）
将 CIFAR-10 转为 png
"""import os
import pickleimport numpy as np
from imageio import imwrite# 数据存放的根目录
base_dir = r'H:\DataStore'
# cifar-10 存放位置
data_dir = os.path.join(base_dir, 'cifar-10-batches-py')
# 训练图片目录
train_dir = os.path.join(base_dir, 'cifar-10-train-png')
# 测试图片目录
test_dir = os.path.join(base_dir, 'cifar-10-test-png')# 这里不进行训练图片的生成
Train = False
Test = True# 反序列化
def unpickle(file_path):with open(file_path, 'rb') as f:_obj = pickle.load(f, encoding='bytes')return _obj# 目录不存在时创建一个
def create_dir(dir_path):if not os.path.isdir(dir_path):os.makedirs(dir_path)def get_label_names():_label_names_obj = unpickle(os.path.join(data_dir, 'batches.meta'))return _label_names_obj[b'label_names']def save_images(i, obj, class_num, label_names, dir_path):# 红、绿、蓝 (channel, width, height)# 一定要使用 b'' 的方式，因为 obj 是 bytes 编码的img = np.reshape(obj[b'data'][i], (3, 32, 32))# 保存为图片使用 (width, height, channel) 格式img = img.transpose(1, 2, 0)# 获取当前图片的类别下标 0-9label_idx = obj[b'labels'][i]# 获取当前图片的名称_label_name: str = label_names[label_idx].decode()train_dir_label_name_path = os.path.join(dir_path, _label_name)create_dir(train_dir_label_name_path)# 图片对应的类别数量+1class_num[label_idx] += 1_image_name = str(class_num[label_idx]) + '.png'image_path = os.path.join(train_dir_label_name_path, _image_name)# 写入图片imwrite(image_path, img)if __name__ == '__main__':_label_names = get_label_names()if Train:# 累计每个类别的数量train_class_num = [0] * 10for i in range(1, 6):data_batch_path = os.path.join(data_dir, 'data_batch_' + str(i))# k: data、labelstrain_batch_obj = unpickle(data_batch_path)print("{} is loading...".format(data_batch_path))# 每个batch中有10000张图片for j in range(0, 10000):save_images(j, train_batch_obj, train_class_num, _label_names, train_dir)print('train loaded')if Test:test_class_num = [0] * 10test_data_path = os.path.join(data_dir, 'test_batch')test_obj = unpickle(test_data_path)for i in range(10000):save_images(i, test_obj, test_class_num, _label_names, test_dir)print('test loaded')