tfds.builder_from_directory

Loads a tfds.core.DatasetBuilder from the given generated dataset path.

tfds.builder_from_directory(
    builder_dir: tfds.typing.PathLike
) -> tfds.core.DatasetBuilder

Reconstructs the tfds.core.DatasetBuilder without requiring the original generation code.

From <builder_dir>/features.json it infers the structure (feature names, nested dict,...) and content (image, sequence,...) of the dataset. The serialization format is defined in tfds.features.FeatureConnector in to_json().

Note: This function only works for datasets generated with TFDS 4.0.0 or above.

Args
builder_dir	str, path of the directory containing the dataset to read ( e.g. ~/tensorflow_datasets/mnist/3.0.0/).

Returns
builder	tfds.core.DatasetBuilder, builder for dataset at the given path.