neuroglancer_precomputed Driver

The neuroglancer_precomputed driver provides access to Neuroglancer Precomputed format volumes backed by any supported Key-Value Storage Layer. It supports reading, writing, and creating new volumes.

Multiscale volumes are supported, but each scale must be opened individually.

json driver/neuroglancer_precomputed : object
Extends:
Required members:
driver : "neuroglancer_precomputed"
kvstore : KvStore | KvStoreUrl

Specifies the underlying storage mechanism.

Optional members:
context : Context

Specifies context resources that augment/override the parent context.

dtype : dtype

Specifies the data type.

rank : integer[0, 32]

Specifies the rank of the TensorStore.

If transform is also specified, the input rank must match. Otherwise, the rank constraint applies to the driver directly.

transform : IndexTransform

Specifies a transform.

schema : Schema

Specifies constraints on the schema.

When opening an existing array, specifies constraints on the existing schema; opening will fail if the constraints do not match. Any soft constraints specified in the chunk_layout are ignored. When creating a new array, a suitable schema will be selected automatically based on the specified schema constraints in combination with any driver-specific constraints.

path : string = ""

Additional path within the KvStore specified by kvstore.

This is joined as an additional "/"-separated path component after any path member directly within kvstore. This is supported for backwards compatibility only; the KvStore.path member should be used instead.

Example

"path/to/data"
open : boolean

Open an existing TensorStore. If neither open nor create is specified, defaults to true.

create : boolean = false

Create a new TensorStore. Specify true for both open and create to permit either opening an existing TensorStore or creating a new TensorStore if it does not already exist.

delete_existing : boolean = false

Delete any existing data at the specified path before creating a new TensorStore. Requires that create is true, and that open is false.

assume_metadata : boolean = false

Neither read nor write stored metadata. Instead, just assume any necessary metadata based on constraints in the spec, using the same defaults for any unspecified metadata as when creating a new TensorStore. The stored metadata need not even exist. Operations such as resizing that modify the stored metadata are not supported. Requires that open is true and delete_existing is false. This option takes precedence over assume_cached_metadata if that option is also specified.

Warning

This option can lead to data corruption if the assumed metadata does not match the stored metadata, or multiple concurrent writers use different assumed metadata.

assume_cached_metadata : boolean = false

Skip reading the metadata when opening. Instead, just assume any necessary metadata based on constraints in the spec, using the same defaults for any unspecified metadata as when creating a new TensorStore. The stored metadata may still be accessed by subsequent operations that need to re-validate or modify the metadata. Requires that open is true and delete_existing is false. The assume_metadata option takes precedence if also specified.

Note

Unlike the assume_metadata option, operations such as resizing that modify the stored metadata are supported (and access the stored metadata).

Warning

This option can lead to data corruption if the assumed metadata does not match the stored metadata, or multiple concurrent writers use different assumed metadata.

cache_pool : ContextResource = "cache_pool"

Cache pool for data.

Specifies or references a previously defined Context.cache_pool. It is normally more convenient to specify a default cache_pool in the context.

metadata_cache_pool : ContextResource

Cache pool for metadata only.

Specifies or references a previously defined Context.cache_pool. If not specified, defaults to the value of cache_pool.

data_copy_concurrency : ContextResource = "data_copy_concurrency"

Specifies or references a previously defined Context.data_copy_concurrency. It is normally more convenient to specify a default data_copy_concurrency in the context.

recheck_cached_metadata : CacheRevalidationBound = "open"

Time after which cached metadata is assumed to be fresh. Cached metadata older than the specified time is revalidated prior to use. The metadata is used to check the bounds of every read or write operation.

Specifying true means that the metadata will be revalidated prior to every read or write operation. With the default value of "open", any cached metadata is revalidated when the TensorStore is opened but is not rechecked for each read or write operation.

recheck_cached_data : CacheRevalidationBound = true

Time after which cached data is assumed to be fresh. Cached data older than the specified time is revalidated prior to being returned from a read operation. Partial chunk writes are always consistent regardless of the value of this option.

The default value of true means that cached data is revalidated on every read. To enable in-memory data caching, you must both specify a cache_pool with a non-zero total_bytes_limit and also specify false, "open", or an explicit time bound for recheck_cached_data.

fill_missing_data_reads = true

Replace missing chunks with the fill value when reading.

If disabled, reading a missing chunk will result in an error. Note that the fill value may still be used when writing a partial chunk. Typically this should only be set to false in the case that store_data_equal_to_fill_value was enabled when writing.

store_data_equal_to_fill_value = false

Store all explicitly written data, even if it is equal to the fill value.

This ensures that explicitly written data, even if it is equal to the fill value, can be distinguished from missing data. If disabled, chunks equal to the fill value may be represented as missing chunks.

scale_index : integer[0, +∞)

Zero-based index of the scale to open or create.

When opening an existing volume, this member may be specified to explicitly indicate the scale to open. Otherwise, the first scale matching the resolution constraint, is chosen. To create a new scale, this must either be left unspecified or equal the number of existing scales (which is also the index that will be assigned to the new scale).

multiscale_metadata : object

Scale-independent metadata.

Specifies the scale-independent metadata of a new volume exactly as in the info file, except that not all members are required. Required when creating a new multiscale volume. When opening an existing volume or creating a new scale within an existing multiscale volume, specifies constraints on the existing metadata.

Optional members:
type : "image" | "segmentation"

Specifying the type of volume.

This has no bearing on TensorStore, but is used by Neuroglancer to pick the default layer type to use. Required when creating a new multiscale volume.

data_type : "uint8" | "uint16" | "uint32" | "uint64" | "float32"

Specifies the data type.

Required when creating a new multiscale volume.

num_channels : integer

Number of channels.

Required when creating a new volume.

scale_metadata : object

Per-scale metadata.

Specifies the per-scale metadata of a new volume as in the info file, except that not all members are required, and the chunk_size member for specifying a single chunk size takes the place of the chunk_sizes member in the info file. Required when creating a new scale. When opening an existing scale, serves to select the scale to open and constrains the existing metadata.

Optional members:
key : string

Specifies the scale key (relative to path).

When opening an existing scale, may be specified to select the scale by key. When creating a new scale, if not specified, defaults to "<xres>_<yres>_<zres>", where [xres, yres, zres] is the resolution.

size : array[3] of integer[0, +∞)

Voxel dimensions of the volume (XYZ order).

Required when creating a new scale if domain is not specified.

Example

[500, 500, 500]
voxel_offset : array[3] of integer = [0, 0, 0]

Voxel origin of the volume (XYZ order).

If specified, size must also be specified.

chunk_size : array[3] of integer[1, +∞)

Chunk dimensions (XYZ order).

When creating a new scale, if not specified explicitly, will be chosen automatically based on the chunk_layout. When opening an existing scale with multiple supported chunk_sizes, may be specified to select the chunk size to use. Otherwise, the first supported chunk size is used.

Example

[500, 500, 500]
resolution : array[3] of number

Voxel size in nanometers (XYZ order).

When opening an existing scale, may be specified to select the scale by resolution.

encoding : "raw" | "jpeg" | "png" | "compressed_segmentation"

Specifies the chunk encoding.

Required when creating a new scale.

jpeg_quality : integer[0, 100] = 75

JPEG encoding quality.

Only applies if encoding is "jpeg". The quality is specified using the IJG (Independent JPEG Group) [0, 100] recommended scale, with 0 having the worst quality (smallest file size) and 100 the best quality (largest file size).

png_level : integer[0, 9]

PNG compression level.

Only applies if encoding is "png". Specifies the zlib compression level between [0, 9], where 0 is uncompressed, with 1 having the fastest compression (largest file size), and 9 the slowest compression (smallest file size). When unset, the library default compression level is used.

compressed_segmentation_block_size : array[3] of number

Block size for compressed segmentation encoding.

If encoding is "compressed_segmentation", required when creating a new scale, and serves as a constraint when opening an existing scale. Otherwise, must not be specified.

sharding : kvstore/neuroglancer_uint64_sharded/ShardingSpec | null

Specifies to use the unsharded (indicated by null) or sharded format. When creating a new scale, if not specified, the unsharded format is used.

Example JSON specifications

Example: Opening the first (or only) scale of an existing multiscale volume.
{
  "driver": "neuroglancer_precomputed",
  "kvstore": "gs://my-bucket/path/to/volume/",
}
Example: Opening an existing scale by index.
{
  "driver": "neuroglancer_precomputed",
  "kvstore": "gs://my-bucket/path/to/volume/",
  "scale_index": 1
}
Example: Opening an existing scale by resolution.
{
  "driver": "neuroglancer_precomputed",
  "kvstore": "gs://my-bucket/path/to/volume/",
  "scale_metadata": {
    "resolution": [4, 4, 40]
  }
}
Example: Opening an existing scale by dimension units.
{
  "driver": "neuroglancer_precomputed",
  "kvstore": "gs://my-bucket/path/to/volume/",
  "schema": {
    "dimension_units": ["4nm", "4nm", "40nm", null]
  }
}
Example: Opening an existing scale by key.
{
  "driver": "neuroglancer_precomputed",
  "kvstore": "gs://my-bucket/path/to/volume/",
  "scale_metadata": {
    "key": "4_4_40"
  }
}
Example: Creating a new scale in an existing multiscale volume.
{
  "driver": "neuroglancer_precomputed",
  "kvstore": "gs://my-bucket/path/to/volume/",
  "scale_metadata": {
    "size": [40000, 50000, 10000],
    "encoding": "compressed_segmentation",
    "compressed_segmentation_block_size": [8, 8, 8],
    "chunk_size": [64, 64, 64],
    "resolution": [8, 8, 40]
  }
}
Example: Creating a new multiscale volume.
{
  "driver": "neuroglancer_precomputed",
  "kvstore": "gs://my-bucket/path/to/volume/",
  "multiscale_metadata": {
    "type": "segmentation",
    "data_type": "uint64",
    "num_channels": 1
  },
  "scale_metadata": {
    "size": [40000, 50000, 10000],
    "encoding": "compressed_segmentation",
    "compressed_segmentation_block_size": [8, 8, 8],
    "chunk_size": [64, 64, 64],
    "resolution": [8, 8, 40]
  }
}

Mapping to TensorStore Schema

Example with unsharded raw encoding

For the following info metadata file:

{
  "@type": "neuroglancer_multiscale_volume",
  "data_type": "uint8",
  "num_channels": 2,
  "scales": [ {
      "chunk_sizes": [[100, 200, 300]],
      "encoding": "raw",
      "key": "8_8_8",
      "resolution": [8.0, 8.0, 8.0],
      "size": [1000, 2000, 3000],
      "voxel_offset": [20, 30, 40]
    }],
  "type": "image"
}

the corresponding Schema is:

{
  "chunk_layout": {
    "grid_origin": [20, 30, 40, 0],
    "inner_order": [3, 2, 1, 0],
    "read_chunk": {"shape": [100, 200, 300, 2]},
    "write_chunk": {"shape": [100, 200, 300, 2]}
  },
  "codec": {"driver": "neuroglancer_precomputed", "encoding": "raw"},
  "dimension_units": [[8.0, "nm"], [8.0, "nm"], [8.0, "nm"], null],
  "domain": {
    "exclusive_max": [1020, 2030, 3040, 2],
    "inclusive_min": [20, 30, 40, 0],
    "labels": ["x", "y", "z", "channel"]
  },
  "dtype": "uint8",
  "rank": 4
}

Example with unsharded compressed segmentation encoding

For the following info metadata file:

{
  "@type": "neuroglancer_multiscale_volume",
  "data_type": "uint64",
  "num_channels": 2,
  "scales": [ {
      "chunk_sizes": [[100, 200, 300]],
      "compressed_segmentation_block_size": [8, 8, 8],
      "encoding": "compressed_segmentation",
      "key": "8_8_8",
      "resolution": [8.0, 8.0, 8.0],
      "size": [1000, 2000, 3000],
      "voxel_offset": [20, 30, 40]
    }],
  "type": "segmentation"
}

the corresponding Schema is:

{
  "chunk_layout": {
    "codec_chunk": {"shape": [8, 8, 8, 1]},
    "grid_origin": [20, 30, 40, 0],
    "inner_order": [3, 2, 1, 0],
    "read_chunk": {"shape": [100, 200, 300, 2]},
    "write_chunk": {"shape": [100, 200, 300, 2]}
  },
  "codec": {"driver": "neuroglancer_precomputed", "encoding": "compressed_segmentation"},
  "dimension_units": [[8.0, "nm"], [8.0, "nm"], [8.0, "nm"], null],
  "domain": {
    "exclusive_max": [1020, 2030, 3040, 2],
    "inclusive_min": [20, 30, 40, 0],
    "labels": ["x", "y", "z", "channel"]
  },
  "dtype": "uint64",
  "rank": 4
}

Example with sharded raw encoding

For the following info metadata file:

{
  "@type": "neuroglancer_multiscale_volume",
  "data_type": "uint8",
  "num_channels": 2,
  "scales": [ {
      "chunk_sizes": [[64, 64, 64]],
      "encoding": "raw",
      "key": "8_8_8",
      "resolution": [8.0, 8.0, 8.0],
      "sharding": {
        "@type": "neuroglancer_uint64_sharded_v1",
        "data_encoding": "gzip",
        "hash": "identity",
        "minishard_bits": 6,
        "minishard_index_encoding": "gzip",
        "preshift_bits": 9,
        "shard_bits": 15
      },
      "size": [34432, 39552, 51508],
      "voxel_offset": [20, 30, 40]
    }],
  "type": "image"
}

the corresponding Schema is:

{
  "chunk_layout": {
    "grid_origin": [20, 30, 40, 0],
    "inner_order": [3, 2, 1, 0],
    "read_chunk": {"shape": [64, 64, 64, 2]},
    "write_chunk": {"shape": [2048, 2048, 2048, 2]}
  },
  "codec": {
    "driver": "neuroglancer_precomputed",
    "encoding": "raw",
    "shard_data_encoding": "gzip"
  },
  "dimension_units": [[8.0, "nm"], [8.0, "nm"], [8.0, "nm"], null],
  "domain": {
    "exclusive_max": [34452, 39582, 51548, 2],
    "inclusive_min": [20, 30, 40, 0],
    "labels": ["x", "y", "z", "channel"]
  },
  "dtype": "uint8",
  "rank": 4
}

Data type

The neuroglancer_precomputed data types map to TensorStore data types of the same name:

Note that internally the neuroglancer_precomputed format always uses little endian encoding.

Domain

The Schema.domain has a rank of 4 and includes both the chunked dimensions as well as the channel dimension. The IndexDomain.shape of the Schema.domain corresponds to driver/neuroglancer_precomputed.scale_metadata.size; IndexDomain.inclusive_min corresponds to driver/neuroglancer_precomputed.scale_metadata.voxel_offset. The channel dimension always has an origin of 0.

The dimension labels are always ["x", "y", "z", "channel"].

The bounds are not resizable.

Example

For the following info metadata file:

{
  "@type": "neuroglancer_multiscale_volume",
  "data_type": "uint8",
  "num_channels": 2,
  "scales": [ {
      "chunk_sizes": [[64, 64, 64]],
      "encoding": "raw",
      "key": "8_8_8",
      "resolution": [8.0, 8.0, 8.0],
      "size": [1000, 2000, 3000],
      "voxel_offset": [20, 30, 40]
    }],
  "type": "image"
}

the corresponding IndexDomain is:

{
  "exclusive_max": [1020, 2030, 3040, 2],
  "inclusive_min": [20, 30, 40, 0],
  "labels": ["x", "y", "z", "channel"]
}

Chunk layout

The ChunkLayout.read_chunk shape, specifying the granularity at which reads may be performed efficiently, corresponds to driver/neuroglancer_precomputed.scale_metadata.chunk_size. The channel dimension is always unchunked, i.e. the chunk size is equal to the number of channels.

The ChunkLayout.grid_origin always matches the origin of the domain.

With respect to the dimension order of [x, y, z, channel], when using the "raw" driver/neuroglancer_precomputed/Codec.encoding, the neuroglancer_precomputed format always stores the data within chunks in colexicographic order (i.e. Fortran order).

When using the unsharded format, the ChunkLayout.write_chunk shape is equal to the ChunkLayout.read_chunk shape.

When using the sharded format, if the sharding parameters are such that each shard corresponds to a rectangular region, then the ChunkLayout.write_chunk shape corresponds to the shape of each shard. Otherwise, the write_chunk shape corresponds to the shape of the entire volume, rounded up to a multiple of the read_chunk shape.

When using the "raw", "png", or "jpeg" driver/neuroglancer_precomputed/Codec.encoding, hard constraints on the ChunkLayout.codec_chunk must not be specified.

When using the "compressed_segmentation" driver/neuroglancer_precomputed/Codec.encoding, the ChunkLayout.codec_chunk shape corresponds to the compressed_segmentation_block_size. Note that the codec chunk size along the channel dimension is always 1.

Selection of chunk layout when creating a new array

When creating a new array, the read chunk shape may be constrained explicitly via ChunkLayout/Grid.shape or implicitly via ChunkLayout/Grid.aspect_ratio and ChunkLayout/Grid.elements. If ChunkLayout/Grid.elements is not specified for the read_chunk, the default is 1 million elements per chunk. A suitable base (read) chunk shape will be chosen based on these constraints.

The write chunk shape may also be constrained separately, either explicitly via ChunkLayout/Grid.shape or implicitly via ChunkLayout/Grid.elements. If no constraints on the write chunk shape are specified, it will be set to match the read chunk shape and the unsharded format will be used. Otherwise, suitable sharding parameters will be chosen automatically to satisfy the constraints. If ChunkLayout/Grid.elements is not specified for the write chunk, the unsharded format is used. Due to the limitations of the sharding format, any ChunkLayout/Grid.aspect_ratio constraints on the write chunk shape are ignored.

When using the "compressed_segmentation" driver/neuroglancer_precomputed/Codec.encoding, the compressed segmentation block size is chosen automatically based on the ChunkLayout.codec_chunk constraints. Both explicit constraints via ChunkLayout/Grid.shape and implicit constraints via ChunkLayout/Grid.aspect_ratio and ChunkLayout/Grid.elements are supported. If ChunkLayout/Grid.elements is not specified, the default is 512 elements per chunk.

Example of unconstrained chunk layout

>>> ts.open(
...     {
...         'driver': 'neuroglancer_precomputed',
...         'kvstore': {
...             'driver': 'memory'
...         }
...     },
...     create=True,
...     dtype=ts.uint16,
...     domain=ts.IndexDomain(
...         inclusive_min=[20, 30, 40, 0],
...         shape=[1000, 2000, 3000, 2],
...     )).result().chunk_layout
ChunkLayout({
  'grid_origin': [20, 30, 40, 0],
  'inner_order': [3, 2, 1, 0],
  'read_chunk': {'shape': [80, 80, 80, 2]},
  'write_chunk': {'shape': [80, 80, 80, 2]},
})

Example of unconstrained chunk layout with compressed segmentation encoding

>>> ts.open(
...     {
...         'driver': 'neuroglancer_precomputed',
...         'kvstore': {
...             'driver': 'memory'
...         }
...     },
...     create=True,
...     dtype=ts.uint32,
...     codec=ts.CodecSpec({
...         'driver': 'neuroglancer_precomputed',
...         'encoding': 'compressed_segmentation',
...     }),
...     domain=ts.IndexDomain(
...         inclusive_min=[20, 30, 40, 0],
...         shape=[1000, 2000, 3000, 2],
...     )).result().chunk_layout
ChunkLayout({
  'codec_chunk': {'shape': [8, 8, 8, 1]},
  'grid_origin': [20, 30, 40, 0],
  'inner_order': [3, 2, 1, 0],
  'read_chunk': {'shape': [80, 80, 80, 2]},
  'write_chunk': {'shape': [80, 80, 80, 2]},
})

Example of chunk layout with separate read and write chunk constraints

>>> ts.open(
...     {
...         'driver': 'neuroglancer_precomputed',
...         'kvstore': {
...             'driver': 'memory'
...         }
...     },
...     create=True,
...     dtype=ts.uint16,
...     chunk_layout=ts.ChunkLayout(
...         chunk_aspect_ratio=[2, 1, 1, 0],
...         read_chunk_elements=2000000,
...         write_chunk_elements=1000000000,
...     ),
...     domain=ts.IndexDomain(
...         inclusive_min=[20, 30, 40, 0],
...         shape=[1000, 2000, 3000, 2],
...     )).result().chunk_layout
ChunkLayout({
  'grid_origin': [20, 30, 40, 0],
  'inner_order': [3, 2, 1, 0],
  'read_chunk': {'shape': [159, 79, 79, 2]},
  'write_chunk': {'shape': [1113, 1264, 632, 2]},
})

Example of chunk layout with explicit chunk shapes

>>> ts.open(
...     {
...         'driver': 'neuroglancer_precomputed',
...         'kvstore': {
...             'driver': 'memory'
...         }
...     },
...     create=True,
...     dtype=ts.uint16,
...     chunk_layout=ts.ChunkLayout(
...         read_chunk_shape=[64, 64, 64, 2],
...         write_chunk_shape=[512, 512, 512, 2],
...     ),
...     domain=ts.IndexDomain(
...         inclusive_min=[20, 30, 40, 0],
...         shape=[1000, 2000, 3000, 2],
...     )).result().chunk_layout
ChunkLayout({
  'grid_origin': [20, 30, 40, 0],
  'inner_order': [3, 2, 1, 0],
  'read_chunk': {'shape': [64, 64, 64, 2]},
  'write_chunk': {'shape': [512, 512, 512, 2]},
})

Codec

Within the Schema.codec, the compression parameters are represented in the same way as in the scale_metadata:

json driver/neuroglancer_precomputed/Codec : object

Neuroglancer Precomputed Codec

Extends:
Required members:
driver : "neuroglancer_precomputed"
Optional members:
encoding : "raw" | "jpeg" | "png" | "compressed_segmentation"

Specifies the chunk encoding.

Required when creating a new scale.

jpeg_quality : integer[0, 100] = 75

JPEG encoding quality.

Only applies if encoding is "jpeg". The quality is specified using the IJG (Independent JPEG Group) [0, 100] recommended scale, with 0 having the worst quality (smallest file size) and 100 the best quality (largest file size).

png_level : integer[0, 9]

PNG compression level.

Only applies if encoding is "png". Specifies the zlib compression level between [0, 9], where 0 is uncompressed, with 1 having the fastest compression (largest file size), and 9 the slowest compression (smallest file size). When unset, the library default compression level is used.

shard_data_encoding : "raw" | "gzip"

Additional data compression when using the sharded format.

This specifies the value of kvstore/neuroglancer_uint64_sharded/ShardingSpec.data_encoding. If not specified, defaults to "gzip" if the encoding is equal to "raw" or "compressed_segmentation", and to "raw" if encoding is equal to "jpeg".

It is an error to specify any other Codec.driver.

Fill value

The neuroglancer_precomputed format does not support specifying a fill value. TensorStore always assumes a fill value of 0.

Dimension Units

The dimension units of the first three (x, y, and z) dimensions always have a base unit of "nm"; the multiplier corresponds to the resolution. It is an error to specify a base unit other than "nm" for these dimensions.

The final (channel) dimension always has an unspecified base unit. It is an error to specify a unit for this dimension.

When creating a new scale, if neither dimension_units nor resolution is specified, a unit of "1nm" is used by default.

When opening an existing scale, the scale to open may be selected based on the specified dimension_units.

Limitations

Resizing is not supported.

Warning

Writing to volumes in the sharded format is supported, but because updating a shard requires rewriting it entirely, write operations may be very inefficient unless special care is taken:

  1. It is most efficient to group writes by shard (i.e. according to the ChunkLayout.write_chunk shape).

  2. The temporary memory required to write a shard is 2 to 3 times the size of the shard. It is therefore advised that the shards be kept as small as possible (while still avoiding an excess number of objects in the underlying key-value store).