MindSpore 数据加载及处理

参考地址：

https://www.mindspore.cn/tutorial/zh-CN/r1.2/dataset.html

========================================================

数据集下载：

mkdir -p ./datasets/MNIST_Data/train ./datasets/MNIST_Data/test
wget -NP ./datasets/MNIST_Data/train https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/train-labels-idx1-ubyte
wget -NP ./datasets/MNIST_Data/train https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/train-images-idx3-ubyte
wget -NP ./datasets/MNIST_Data/test https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/t10k-labels-idx1-ubyte
wget -NP ./datasets/MNIST_Data/test https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/t10k-images-idx3-ubyte
tree ./datasets/MNIST_Data

顺序读取N个样本：

import mindspore.dataset as ds
from mindspore import dtype as mstype

DATA_DIR = "./datasets/MNIST_Data/train"
sampler = ds.SequentialSampler(num_samples=3)
dataset = ds.MnistDataset(DATA_DIR, sampler=sampler)

for data in dataset.create_dict_iterator():
    print("Image shape: {}".format(data['image'].shape), ", Label: {}".format(data['label']))
    print(data['label'], "\t", data['label'].astype(mstype.float32))

自定义数据集

import mindspore.dataset as ds

import numpy as np

np.random.seed(58)

class DatasetGenerator:
    def __init__(self):
        self.data = np.random.sample((5, 2))
        self.label = np.random.sample((5, 1))

    def __getitem__(self, index):
        return self.data[index], self.label[index]

    def __len__(self):
        return len(self.data)

dataset_generator = DatasetGenerator()
dataset = ds.GeneratorDataset(dataset_generator, ["data", "label"], shuffle=False)

for i, data in enumerate(dataset.create_dict_iterator()):
    print("第 %d 个样本"%i)
    print('{}'.format(data["data"]), '{}'.format(data["label"]))

对自定义数据集进行一定预处理：

import mindspore.dataset as ds

import numpy as np

np.random.seed(58)

class DatasetGenerator:
    def __init__(self):
        self.data = np.random.sample((5, 2))
        self.label = np.random.sample((5, 1))

    def __getitem__(self, index):
        return self.data[index], self.label[index]

    def __len__(self):
        return len(self.data)

dataset_generator = DatasetGenerator()
dataset = ds.GeneratorDataset(dataset_generator, ["data", "label"], shuffle=False)

# 随机打乱数据顺序
dataset = dataset.shuffle(buffer_size=10)
# 对数据集进行分批
dataset = dataset.batch(batch_size=2)

for i, data in enumerate(dataset.create_dict_iterator()):
    print("第 %d 次选取样本"%i)
    print("data: \n{}".format(data["data"]))
    print("label: \n{}".format(data["label"]))

数据处理及增强

import matplotlib.pyplot as plt

import mindspore.dataset as ds
from mindspore.dataset.vision import Inter
import mindspore.dataset.vision.c_transforms as c_vision

DATA_DIR = './datasets/MNIST_Data/train'
_number_samples = 3

mnist_dataset = ds.MnistDataset(DATA_DIR, num_samples=_number_samples, shuffle=False)

resize_op = c_vision.Resize(size=(200,200), interpolation=Inter.LINEAR)
crop_op = c_vision.RandomCrop(150)     # 随机将图像裁剪成150尺寸
transforms_list = [resize_op, crop_op]
mnist_dataset = mnist_dataset.map(operations=transforms_list, input_columns=["image"])

# 查看数据原图
mnist_it = mnist_dataset.create_dict_iterator()

for _ in range(_number_samples):
    data = next(mnist_it)
    print(type(data['image']), data['image'].shape)

    plt.imshow(data['image'].asnumpy().squeeze(), cmap=plt.cm.gray)
    plt.title(data['label'].asnumpy(), fontsize=20)
    plt.show()