Dataset Usage¶

The high-level dataset API allows to interact with datasets while automatically maintaining meta data for any dataset, such as the datasource-properties.json.

The Dataset class is the entry-point for this API. The dataset stores the data on disk in .wkw-files.

Each dataset consists of one or more layers, which themselves can comprise multiple magnifications represented via MagViews.

import numpy as np

import webknossos as wk

# ruff: noqa: F841 unused-variable


def main() -> None:
    #####################
    # Opening a dataset #
    #####################

    dataset = wk.Dataset.open("testdata/simple_wkw_dataset")
    # Assuming that the dataset has a layer "color"
    # and the layer has the magnification 1
    layer = dataset.get_layer("color")
    mag1 = layer.get_mag("1")

    ######################
    # Creating a dataset #
    ######################

    dataset = wk.Dataset("testoutput/my_new_dataset", voxel_size=(1, 1, 1))
    layer = dataset.add_layer(
        layer_name="color",
        category="color",
        dtype_per_channel="uint8",
        num_channels=3,
        bounding_box=wk.BoundingBox((10, 20, 30), (512, 512, 32)),
    )
    mag1 = layer.add_mag("1")
    mag2 = layer.add_mag("2")

    ##########################
    # Writing into a dataset #
    ##########################

    # The properties are updated automatically
    # when the written data exceeds the bounding box in the properties
    mag1.write(
        absolute_offset=(10, 20, 30),
        # assuming the layer has 3 channels:
        data=(np.random.rand(3, 512, 512, 32) * 255).astype(np.uint8),
        allow_unaligned=True,
    )

    mag2.write(
        absolute_offset=(10, 20, 30),
        data=(np.random.rand(3, 256, 256, 16) * 255).astype(np.uint8),
        allow_unaligned=True,
    )

    ##########################
    # Reading from a dataset #
    ##########################

    data_in_mag1 = mag1.read()  # the offset and size from the properties are used
    data_in_mag1_subset = mag1.read(absolute_offset=(10, 20, 30), size=(512, 512, 32))

    data_in_mag2 = mag2.read()
    data_in_mag2_subset = mag2.read(absolute_offset=(10, 20, 30), size=(512, 512, 32))
    assert data_in_mag2_subset.shape == (3, 256, 256, 16)

    #####################
    # Copying a dataset #
    #####################

    copy_of_dataset = dataset.copy_dataset(
        "testoutput/copy_of_dataset",
        chunk_shape=(8, 8, 8),
        shard_shape=(64, 64, 64),
        compress=True,
    )
    new_layer = dataset.add_layer(
        layer_name="segmentation",
        category="segmentation",
        dtype_per_channel="uint8",
        largest_segment_id=0,
    )
    # Link a layer of the initial dataset to the copy:
    sym_layer = copy_of_dataset.add_symlink_layer(new_layer)


if __name__ == "__main__":
    main()

Parallel Access of WEBKNOSSOS Datasets¶

Please consider these restrictions when accessing a WEBKNOSSOS dataset in a multiprocessing-context:

When writing shards in parallel, json_update_allowed should be set to False to disable the automatic update of the bounding box metadata. Otherwise, race conditions may happen. The user is responsible for updating the bounding box manually.
When writing to chunks in shards, one chunk may only be written to by one actor at any time.
When writing to compressed shards, one shard may only be written to by one actor at any time.
For Zarr datasets, parallel write access to shards is not allowed at all.
Reading in parallel without concurrent writes is fine.