from python.tests.test_stream import store_path

Acquire Zarr streaming library

This library supports chunked, compressed, multiscale streaming to Zarr, with OME-NGFF metadata.

Building

Installing dependencies

This library has the following dependencies:

• c-blosc v1.21.5
• nlohmann-json v3.11.3
• minio-cpp v0.3.0

We use vcpkg to install them, as it integrates well with CMake. To install vcpkg, clone the repository and bootstrap it:

git clone https://github.com/microsoft/vcpkg.git
cd vcpkg && ./bootstrap-vcpkg.sh

and then add the vcpkg directory to your path. If you are using bash, you can do this by running the following snippet from the vcpkg/ directory:

cat >> ~/.bashrc <<EOF
export VCPKG_ROOT=${PWD}
export PATH=\$VCPKG_ROOT:\$PATH
EOF

If you're using Windows, learn how to set environment variables here. You will need to set both the VCPKG_ROOT and PATH variables in the system control panel.

Configuring

To build the library, you can use CMake:

cmake --preset=default -B /path/to/build /path/to/source

On Windows, you'll need to specify the target triplet to ensure that all dependencies are built as static libraries:

cmake --preset=default -B /path/to/build -DVCPKG_TARGET_TRIPLET=x64-windows-static /path/to/source

Aside from the usual CMake options, you can choose to disable tests by setting BUILD_TESTING to OFF:

cmake --preset=default -B /path/to/build -DBUILD_TESTING=OFF /path/to/source

To build the Python bindings, make sure pybind11 is installed. Then, you can set BUILD_PYTHON to ON:

cmake --preset=default -B /path/to/build -DBUILD_PYTHON=ON /path/to/source

Building

After configuring, you can build the library:

cmake --build /path/to/build

Installing for Python

To install the Python bindings, you can run:

pip install .

Note

It is highly recommended to use virtual environments for Python, e.g. using venv or conda. In this case, make sure pybind11 is installed in this environment, and that the environment is activated before installing the bindings.

Usage

The library provides two main interfaces. First, ZarrStream, representing an output stream to a Zarr dataset. Second, ZarrStreamSettings to configure a Zarr stream.

A typical use case for a 4-dimensional acquisition might look like this:

ZarrStreamSettings settings = (ZarrStreamSettings){
    .store_path = "my_stream.zarr",
    .data_type = ZarrDataType_uint16,
    .version = ZarrVersion_3,
};
settings.store_path = "my_stream.zarr";
settings.data_type = ZarrDataType_uint16;
settings.version = ZarrVersion_3;

ZarrStreamSettings_create_dimension_array(&settings, 4);
settings.dimensions[0] = (ZarrDimensionProperties){
    .name = "t",
    .type = ZarrDimensionType_Time,
    .array_size_px = 0,      // this is the append dimension
    .chunk_size_px = 100,    // 100 time points per chunk
    .shard_size_chunks = 10, // 10 chunks per shard
};

settings.dimensions[1] = (ZarrDimensionProperties){
    .name = "c",
    .type = ZarrDimensionType_Channel,
    .array_size_px = 3,     // 3 channels
    .chunk_size_px = 1,     // 1 channel per chunk
    .shard_size_chunks = 1, // 1 chunk per shard
};

settings.dimensions[2] = (ZarrDimensionProperties){
    .name = "y",
    .type = ZarrDimensionType_Space,
    .array_size_px = 1080,  // height
    .chunk_size_px = 270,   // 4 x 4 tiles of size 270 x 480
    .shard_size_chunks = 2, // 2 x 2 tiles per shard
};

settings.dimensions[3] = (ZarrDimensionProperties){
    .name = "x",
    .type = ZarrDimensionType_Space,
    .array_size_px = 1920,  // width
    .chunk_size_px = 480,   // 4 x 4 tiles of size 270 x 480
    .shard_size_chunks = 2, // 2 x 2 tiles per shard
};

ZarrStream* stream = ZarrStream_create(&settings);

size_t bytes_written;
ZarrStream_append(stream, my_frame_data, my_frame_size, &bytes_written);
assert(bytes_written == my_frame_size);

Look at acquire.zarr.h for more details.

This acquisition in Python would look like this:

import acquire_zarr as aqz
import numpy as np

settings = aqz.StreamSettings(
    store_path="my_stream.zarr",
    data_type=aqz.DataType.UINT16,
    version=aqz.ZarrVersion.V3
)

settings.dimensions.extend([
    aqz.Dimension(
        name="t",
        type=aqz.DimensionType.TIME,
        array_size_px=0,
        chunk_size_px=100,
        shard_size_chunks=10
    ),
    aqz.Dimension(
        name="c",
        type=aqz.DimensionType.CHANNEL,
        array_size_px=3,
        chunk_size_px=1,
        shard_size_chunks=1
    ),
    aqz.Dimension(
        name="y",
        type=aqz.DimensionType.SPACE,
        array_size_px=1080,
        chunk_size_px=270,
        shard_size_chunks=2
    ),
    aqz.Dimension(
        name="x",
        type=aqz.DimensionType.SPACE,
        array_size_px=1920,
        chunk_size_px=480,
        shard_size_chunks=2
    )
])

# Generate some random data: one time point, all channels, full frame
my_frame_data = np.random.randint(0, 2**16, (3, 1080, 1920), dtype=np.uint16)

stream = aqz.ZarrStream(settings)
stream.append(my_frame_data)