tfds-blog-post-img.py

import tensorflow_datasets as tfds

# Download the dataset and create a tf.data.Dataset
ds, info = tfds.load("mnist", split="train", with_info=True)

# Access relevant metadata with DatasetInfo
print(info.splits["train"].num_examples)
print(info.features["label"].num_classes)

# Build your input pipeline
ds = ds.batch(128).repeat(10)

# And get NumPy arrays if you'd like
for ex in tfds.as_numpy(ds):
  np_image, np_label = ex["image"], ex["label"]