MNIST dataset from torchvision versus keras.datasets

[dgcovell]

I am trying to implement the code at

https://github.com/tschechlovdev/AutoEncoder_KMeans/blob/main/AutoEncoder_KMeans_MNIST.ipynb

but I would like to replace the mnist dataset with my own data. From the above site the steps for loading MNIST are

from torchvision.datasets import MNIST
from torch.utils.data import ConcatDataset
from torchvision import transforms
transform = transforms.Compose([transforms.ToTensor(),
transforms.Normalize((0.5,), (0.5,)),
])
trainset = MNIST('./', download=True,
train=True,
transform=transform)
testset = MNIST('./', download=True,
train=False,
transform=transform)

Alternatively the mnist dataset can be loaded with
from keras.datasets import mnist

#loading the dataset
(train_X, train_y), (test_X, test_y) = mnist.load_data()

My question is how to get from the mnist dataset to the format used for the MNIST dataset. I see that the latter has transformation and normalization within the transform step. Transform apparently converts the mnist (28x28) to a 784 vector.
Ideally export/importing the
MNIST train/testsets to a csv file would be helpful to me. However the MNIST test/trainsets are not compatible for creating a dataframe (df_testset = pd.DataFrame(testset) yields the error ValueError: DataFrame constructor not properly called!).

If anyone could provide the steps connecting the MNIST and mnist datasets that would be appreciated. My overall goal is to use the autoencoder utilities in this blog to process my own data.

Thanks