Multiscale datasets via python

[TillLeiden]

I have the following question.
Is there a possibility via python to access a local multiscale dataset? Maybe I missed it.
Something like

import argparse
import time

import neuroglancer
import neuroglancer.cli
import numpy as np


def main():
    parser = argparse.ArgumentParser()
    neuroglancer.cli.add_server_arguments(parser)
    args = parser.parse_args()

    neuroglancer.cli.handle_server_arguments(args)
    neuroglancer.set_server_bind_address(bind_address="localhost", bind_port=9999)
    viewer = neuroglancer.Viewer()

    # load dataset / create dummy multiscale data
    data = np.random.randint(0, 255, size=(256, 256, 256), dtype=np.uint8)

    multiscale_data = []
    coordinate_spaces = []
    for i in range(4):
        multiscale_data += [data[:: 2**i, :: 2**i, :: 2**i]]
        coordinate_spaces += [
            neuroglancer.CoordinateSpace(
                names=["x", "y", "z"],
                units=["nm", "nm", "nm"],
                scales=[2**i, 2**i, 2**i],
            )
        ]

    # set up visualization
    with viewer.txn() as state:
        # # load single dataset
        # state.layers["image"] = neuroglancer.ImageLayer(
        #     source=neuroglancer.LocalVolume(
        #         data=multiscale_data[0],
        #         volume_type="image",
        #         voxel_offset=[0, 0, 0],
        #         dimensions=coordinate_spaces[0],
        #     ),
        # )

        # interesting, but correct
        state.layers["image"] = neuroglancer.ImageLayer(
            source=[
                neuroglancer.LocalVolume(
                    data=data,
                    volume_type="image",
                    voxel_offset=[0, 0, 0],
                    dimensions=cs,
                )
                for data, cs in zip(multiscale_data, coordinate_spaces)
            ],
        )

    # show url
    print(viewer)

    # keep viewer alive
    while True:
        time.sleep(1)


if __name__ == "__main__":
    main()

We have huge datasets (up to petabytes) stored in hdf5 files, similar to a precomputed format, from which a fast 3d visualization should be possible. I guess another solution would be, to write something like a tile server for our format, which neuroglancer can access.