import pickle
import numpy as np
from mindspore.dataset import GeneratorDataset
def setup():
    def load(fileName: str):
        with open(file=fileName, mode="rb") as handler:
            data = pickle.load(file=handler, encoding="latin1")
        return data
    trains = load("D:/dataset/cifar-100-python/train")
    tests = load("D:/dataset/cifar-100-python/test")
    metas = load("D:/dataset/cifar-100-python/meta")
    return trains, tests, metas

def train():
    trains, tests, metas = setup()
    print(trains["data"][0],trains["coarse_labels"][0],trains["fine_labels"][0] )
    print(len(trains["coarse_labels"]))
    for key in trains.keys():
        print(f"key = {key}, len(trains[key]) = {len(trains[key])}")
    print("--------------------------------------------------")
    for key in tests.keys():
        print(f"key = {key}, len(tests[key]) = {len(tests[key])}")
    print("--------------------------------------------------")
    for key in metas.keys():
        print(f"key = {key}, len(metas[key]) = {len(metas[key])}")

#
class LoadCifar100Dataset():
    def __init__(self,type):
        self.type = type
        self.trains, self.tests = self.setup()
    def __len__(self):
        if self.type == "train":
            return len(self.trains["coarse_labels"])
        else:
            return len(self.tests["coarse_labels"])
    def __getitem__(self, idx):
        if self.type=="train":
            # return {
            #     "data":np.reshape(self.trains["data"][idx].astype(np.float32), (3,32,32)),
            #     "coarse_labels":self.trains["coarse_labels"][idx],
            #     "fine_labels":self.trains["fine_labels"][idx]
            # }
            #返回的是元组
            return (np.reshape(self.trains["data"][idx].astype(np.float32), (3,32,32)), self.trains["coarse_labels"][idx], self.trains["fine_labels"][idx])
        else:
            return (np.reshape(self.tests["data"][idx].astype(np.float32), (3,32,32)), self.tests["coarse_labels"][idx], self.tests["fine_labels"][idx])

    def setup(self):
        def load(fileName: str):
            with open(file=fileName, mode="rb") as handler:
                data = pickle.load(file=handler, encoding="latin1")
            return data

        trains = load("D:/dataset/cifar-100-python/train")
        tests = load("D:/dataset/cifar-100-python/test")
        return trains, tests

if __name__ == '__main__':
    # train()
    cifar100_train_dataset = LoadCifar100Dataset("train")
    cifar100_test_dataset = LoadCifar100Dataset("test")
    train_generator = GeneratorDataset(cifar100_train_dataset,column_names=["image","coarse_labels","fine_labels"])
    train_generator = train_generator.batch(10)
    size = train_generator.get_dataset_size()
    print(size)
    for i, (imgs, coarse_labels, fine_labels) in enumerate(train_generator.create_tuple_iterator()):
        print(imgs.shape)

数据集下载：

 浏览器打开：https://www.cs.toronto.edu/~kriz/cifar-100-python.tar.gz 进行下载至本地。