From other inputs (generator, SQL database, range, and others)

Although we will not list them all, it is good to keep in mind that tf.data.Dataset can be defined from a wide range of input sources. For example, datasets simply iterating over numbers can be initialized with the .range() static method. Datasets can also be built upon Python generators with .from_generator(). Finally, even if elements are stored in a SQL database, TensorFlow provides some (experimental) tools to query it, including the following:

dataset = tf.data.experimental.SqlDataset(
    "sqlite", "/path/to/my_db.sqlite3",
    "SELECT img_filename, label FROM images", (tf.string, tf.int32))

For more specific dataset instantiators, we invite our readers to check the tf.data documentation.