spine.io.augment

Data augmentation managers and modules.

class spine.io.augment.AugmentManager(**augmenters: Mapping[str, Any] | None)

Generic class to handle ordered data augmentation modules.

Methods

__call__(data)	Augment the data products in one event.
copy_meta(meta)	Return a detached copy of the metadata.

static copy_meta(meta: ImageMeta3D) → ImageMeta3D

Return a detached copy of the metadata.

Parameters:

meta (Meta) – Metadata object to copy

Returns:

Detached metadata copy

Return type:

Meta

spine.io.augment.Augmenter

alias of AugmentManager

class spine.io.augment.AugmentBase

Base class for augmentation modules.

Methods

__call__(data, meta, keys, context)	Apply an augmentation module.
apply(data, meta, keys, context)	Apply an augmentation to one event.
cm_to_voxel(coords_cm, meta, dtype)	Convert detector coordinates at voxel centers back to indices.
make_grid_aligned_meta(meta, lower_bound, ...)	Build a box metadata object snapped to the source voxel grid.
make_snapped_meta(meta, size, count, lower)	Build metadata snapped to the source grid from a proposed lower edge.
parse_optional_vector(value, name)	Parse an optional scalar-or-vector parameter into a length-3 array.
resolve_activity_center(data, keys, meta[, ...])	Estimate an activity center from all coordinate-carrying tensors.
resolve_activity_stats(data, keys, meta[, ...])	Estimate activity center and spread from coordinate-carrying tensors.
resolve_center(meta[, center, use_geo_center])	Resolve the pivot center for a geometric transform.
sample_box_lower(lower, upper, dimensions[, ...])	Sample the lower corner of a crop/mask box.
voxel_to_cm(coords, meta)	Convert voxel indices to detector coordinates at voxel centers.

name = ''

abstractmethod apply(data: dict[str, Any], meta: ImageMeta3D, keys: list[str], context: dict[str, Any]) → tuple[dict[str, Any], ImageMeta3D]

Apply an augmentation to one event.

Parameters:

• data (dict) – Dictionary of event data products to augment

• meta (Meta) – Shared image metadata

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• context (dict) – Shared augmentation context built by the manager

Returns:

Updated data dictionary and shared metadata

Return type:

Tuple[Dict[str, Any], Meta]

static resolve_center(meta: ImageMeta3D, center: ndarray | None = None, use_geo_center: bool = False) → ndarray

Resolve the pivot center for a geometric transform.

Parameters:

• meta (Meta) – Current image metadata

• center (np.ndarray, optional) – Explicit center in detector coordinates (cm)

• use_geo_center (bool, default False) – If True, use the detector TPC center from the geometry manager

Returns:

(3,) Pivot center in detector coordinates (cm)

Return type:

np.ndarray

static voxel_to_cm(coords: ndarray, meta: ImageMeta3D) → ndarray

Convert voxel indices to detector coordinates at voxel centers.

Parameters:

• coords (np.ndarray) – (N, 3) Array of voxel indices

• meta (Meta) – Metadata used to convert voxel indices to detector coordinates

Returns:

(N, 3) Detector coordinates in cm at voxel centers

Return type:

np.ndarray

static cm_to_voxel(coords_cm: ndarray, meta: ImageMeta3D, dtype: dtype) → ndarray

Convert detector coordinates at voxel centers back to indices.

Parameters:

• coords_cm (np.ndarray) – (N, 3) Detector coordinates in cm at voxel centers

• meta (Meta) – Metadata used to convert detector coordinates back to pixel space

• dtype (np.dtype) – Output dtype to use for the returned voxel indices

Returns:

(N, 3) Array of voxel indices

Return type:

np.ndarray

static parse_optional_vector(value: float | list[float] | tuple[float, ...] | ndarray | None, name: str) → ndarray | None

Parse an optional scalar-or-vector parameter into a length-3 array.

Parameters:

• value (float or sequence or np.ndarray, optional) – Input value to parse. Scalars are broadcast to all three axes.

• name (str) – Parameter name used in validation error messages.

Returns:

Length-3 vector if a value is provided, otherwise None

Return type:

np.ndarray or None

static resolve_activity_center(data: dict[str, Any], keys: list[str], meta: ImageMeta3D, weighted: bool = False, feature_index: int = 0) → ndarray

Estimate an activity center from all coordinate-carrying tensors.

Parameters:

• data (dict) – Dictionary of event data products

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• meta (Meta) – Shared image metadata

• weighted (bool, default False) – If True, weight the center by the absolute feature value in the requested feature column

• feature_index (int, default 0) – Feature column to use when weighted=True

Returns:

(3,) Activity center in detector coordinates (cm)

Return type:

np.ndarray

static resolve_activity_stats(data: dict[str, Any], keys: list[str], meta: ImageMeta3D, weighted: bool = False, feature_index: int = 0) → tuple[ndarray, ndarray | None]

Estimate activity center and spread from coordinate-carrying tensors.

Parameters:

• data (dict) – Dictionary of event data products

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• meta (Meta) – Shared image metadata

• weighted (bool, default False) – If True, weight the center and spread by the absolute feature value in the requested feature column

• feature_index (int, default 0) – Feature column to use when weighted=True

Returns:

(3,) Activity center and standard deviation in detector coordinates (cm). If no activity is available, the center falls back to the metadata center and the spread is None.

Return type:

Tuple[np.ndarray, np.ndarray or None]

static sample_box_lower(lower: ndarray, upper: ndarray, dimensions: ndarray, anchor: ndarray | None = None, spread: ndarray | None = None) → ndarray

Sample the lower corner of a crop/mask box.

Parameters:

• lower (np.ndarray) – Lower detector bounds of the allowed sampling region in cm

• upper (np.ndarray) – Upper detector bounds of the allowed sampling region in cm

• dimensions (np.ndarray) – Requested crop or mask box dimensions in cm

• anchor (np.ndarray, optional) – Preferred box center in cm. If provided, sampling is biased around this center.

• spread (np.ndarray, optional) – Standard deviation of the Gaussian proposal in cm when sampling around an anchor. If not provided, a fraction of the available range is used.

Returns:

• np.ndarray – Lower detector corner of the sampled box in cm

• If an anchor is provided, sample the box center around it with a normal

• distribution and clamp to the valid range. Otherwise use a uniform draw.

static make_grid_aligned_meta(meta: ImageMeta3D, lower_bound: ndarray, upper_bound: ndarray, count: ndarray, sampled_lower: ndarray) → ImageMeta3D

Build a box metadata object snapped to the source voxel grid.

Parameters:

• meta (Meta) – Reference metadata that defines the source voxel grid

• lower_bound (np.ndarray) – Minimum allowed lower edge in detector coordinates (cm)

• upper_bound (np.ndarray) – Maximum allowed upper edge in detector coordinates (cm)

• count (np.ndarray) – Requested number of voxels along each axis

• sampled_lower (np.ndarray) – Proposed lower edge in detector coordinates (cm)

Returns:

Grid-aligned metadata for the sampled box

Return type:

Meta

static make_snapped_meta(meta: ImageMeta3D, size: ndarray, count: ndarray, lower: ndarray) → ImageMeta3D

Build metadata snapped to the source grid from a proposed lower edge.

Parameters:

• meta (Meta) – Reference metadata that defines the source voxel grid

• size (np.ndarray) – Pixel size for the transformed metadata

• count (np.ndarray) – Pixel counts for the transformed metadata

• lower (np.ndarray) – Proposed lower edge in detector coordinates (cm)

Returns:

Grid-aligned metadata for the transformed image volume

Return type:

Meta

class spine.io.augment.CropAugment(min_dimensions: ndarray | None = None, max_dimensions: ndarray | None = None, lower: ndarray | None = None, upper: ndarray | None = None, use_geo_boundaries: bool = False, center_mode: str = 'uniform', center_spread: ndarray | None = None, center_feature_index: int = 0, active_volume: bool = False, keep_meta: bool = True)

Generic class to handle cropping images.

Methods

__call__(data, meta, keys, context)	Apply an augmentation module.
active_volume_mask(coords_cm)	Check which detector coordinates lie inside any module active volume.
apply(data, meta, keys, context)	Randomly crop the image within the pre-defined range.
cm_to_voxel(coords_cm, meta, dtype)	Convert detector coordinates at voxel centers back to indices.
generate_active_volume_meta(meta)	Generate metadata aligned to the detector active-volume envelope.
generate_crop(data, meta, keys)	Generate crop box metadata to apply to voxel index sets.
make_grid_aligned_meta(meta, lower_bound, ...)	Build a box metadata object snapped to the source voxel grid.
make_snapped_meta(meta, size, count, lower)	Build metadata snapped to the source grid from a proposed lower edge.
parse_optional_vector(value, name)	Parse an optional scalar-or-vector parameter into a length-3 array.
resolve_activity_center(data, keys, meta[, ...])	Estimate an activity center from all coordinate-carrying tensors.
resolve_activity_stats(data, keys, meta[, ...])	Estimate activity center and spread from coordinate-carrying tensors.
resolve_center(meta[, center, use_geo_center])	Resolve the pivot center for a geometric transform.
sample_box_lower(lower, upper, dimensions[, ...])	Sample the lower corner of a crop/mask box.
voxel_to_cm(coords, meta)	Convert voxel indices to detector coordinates at voxel centers.

name = 'crop'

apply(data: dict[str, Any], meta: ImageMeta3D, keys: list[str], context: dict[str, Any]) → tuple[dict[str, Any], ImageMeta3D]

Randomly crop the image within the pre-defined range.

Parameters:

• data (dict) – Dictionary of event data products to augment

• meta (Meta) – Shared image metadata

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• context (dict) – Shared augmentation context

Returns:

Updated data dictionary and cropped metadata

Return type:

Tuple[Dict[str, Any], Meta]

active_volume_mask(coords_cm: ndarray) → ndarray

Check which detector coordinates lie inside any module active volume.

Parameters:

coords_cm (np.ndarray) – (N, 3) Detector coordinates in cm

Returns:

Boolean mask selecting coordinates inside at least one module box

Return type:

np.ndarray

generate_active_volume_meta(meta: ImageMeta3D) → ImageMeta3D

Generate metadata aligned to the detector active-volume envelope.

Parameters:

meta (Meta) – Metadata of the current image volume

Returns:

Metadata covering the overlap between the current image grid and the detector active-volume envelope

Return type:

Meta

generate_crop(data: dict[str, Any], meta: ImageMeta3D, keys: list[str]) → ImageMeta3D

Generate crop box metadata to apply to voxel index sets.

Parameters:

• data (dict) – Dictionary of event data products used to estimate an activity center when activity-biased sampling is enabled

• meta (Meta) – Metadata of the original image

• keys (List[str]) – Keys corresponding to data products that carry coordinates

Returns:

Metadata describing the cropped image volume

Return type:

Meta

class spine.io.augment.FlipAugment(axis: int, center: ndarray | None = None, use_geo_center: bool = False, keep_meta: bool = True, p: float = 1.0)

Reflect voxel coordinates across a plane normal to one detector axis.

Methods

__call__(data, meta, keys, context)	Apply an augmentation module.
apply(data, meta, keys, context)	Reflect coordinates across the requested plane.
cm_to_voxel(coords_cm, meta, dtype)	Convert detector coordinates at voxel centers back to indices.
flip_points(points, pivot)	Reflect detector coordinates across the configured plane.
generate_meta(meta, pivot)	Generate metadata for the reflected image.
make_grid_aligned_meta(meta, lower_bound, ...)	Build a box metadata object snapped to the source voxel grid.
make_snapped_meta(meta, size, count, lower)	Build metadata snapped to the source grid from a proposed lower edge.
parse_optional_vector(value, name)	Parse an optional scalar-or-vector parameter into a length-3 array.
resolve_activity_center(data, keys, meta[, ...])	Estimate an activity center from all coordinate-carrying tensors.
resolve_activity_stats(data, keys, meta[, ...])	Estimate activity center and spread from coordinate-carrying tensors.
resolve_center(meta[, center, use_geo_center])	Resolve the pivot center for a geometric transform.
sample_box_lower(lower, upper, dimensions[, ...])	Sample the lower corner of a crop/mask box.
voxel_to_cm(coords, meta)	Convert voxel indices to detector coordinates at voxel centers.

name = 'flip'

apply(data: dict[str, Any], meta: ImageMeta3D, keys: list[str], context: dict[str, Any]) → tuple[dict[str, Any], ImageMeta3D]

Reflect coordinates across the requested plane.

Parameters:

• data (dict) – Dictionary of event data products to reflect

• meta (Meta) – Shared image metadata before reflection

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• context (dict) – Shared augmentation context

Returns:

Updated data dictionary and reflected metadata

Return type:

Tuple[Dict[str, Any], Meta]

flip_points(points: ndarray, pivot: ndarray) → ndarray

Reflect detector coordinates across the configured plane.

Parameters:

• points (np.ndarray) – (N, 3) Detector coordinates in cm

• pivot (np.ndarray) – (3,) Point on the reflection plane in detector coordinates (cm)

Returns:

(N, 3) Reflected detector coordinates in cm

Return type:

np.ndarray

generate_meta(meta: ImageMeta3D, pivot: ndarray) → ImageMeta3D

Generate metadata for the reflected image.

Parameters:

• meta (Meta) – Metadata of the image before reflection

• pivot (np.ndarray) – (3,) Point on the reflection plane in detector coordinates (cm)

Returns:

Metadata of the reflected image volume

Return type:

Meta

class spine.io.augment.JitterAugment(max_offset: int | tuple[int, int, int] | list[int] | ndarray, distribution: str = 'uniform', poisson_lambda: float | tuple[float, float, float] | list[float] | ndarray = 1.0, clip: bool = True)

Generic class to handle voxel coordinate jitter.

Methods

__call__(data, meta, keys, context)	Apply an augmentation module.
apply(data, meta, keys, context)	Apply per-voxel coordinate jitter.
cm_to_voxel(coords_cm, meta, dtype)	Convert detector coordinates at voxel centers back to indices.
generate_offsets(num_voxels)	Generate random integer voxel offsets.
generate_poisson_offsets(num_voxels)	Generate signed Poisson-distributed voxel offsets.
make_grid_aligned_meta(meta, lower_bound, ...)	Build a box metadata object snapped to the source voxel grid.
make_snapped_meta(meta, size, count, lower)	Build metadata snapped to the source grid from a proposed lower edge.
parse_optional_vector(value, name)	Parse an optional scalar-or-vector parameter into a length-3 array.
resolve_activity_center(data, keys, meta[, ...])	Estimate an activity center from all coordinate-carrying tensors.
resolve_activity_stats(data, keys, meta[, ...])	Estimate activity center and spread from coordinate-carrying tensors.
resolve_center(meta[, center, use_geo_center])	Resolve the pivot center for a geometric transform.
sample_box_lower(lower, upper, dimensions[, ...])	Sample the lower corner of a crop/mask box.
voxel_to_cm(coords, meta)	Convert voxel indices to detector coordinates at voxel centers.

name = 'jitter'

apply(data: dict[str, Any], meta: ImageMeta3D, keys: list[str], context: dict[str, Any]) → tuple[dict[str, Any], ImageMeta3D]

Apply per-voxel coordinate jitter.

Parameters:

• data (dict) – Dictionary of event data products to augment

• meta (Meta) – Shared image metadata

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• context (dict) – Shared augmentation context

Returns:

Updated data dictionary and unchanged metadata

Return type:

Tuple[Dict[str, Any], Meta]

generate_offsets(num_voxels: int) → ndarray

Generate random integer voxel offsets.

Parameters:

num_voxels (int) – Number of voxel coordinates to jitter

Returns:

Integer per-voxel offsets of shape (num_voxels, 3)

Return type:

np.ndarray

generate_poisson_offsets(num_voxels: int) → ndarray

Generate signed Poisson-distributed voxel offsets.

Parameters:

num_voxels (int) – Number of voxel coordinates to jitter

Returns:

Integer per-voxel offsets of shape (num_voxels, 3)

Return type:

np.ndarray

class spine.io.augment.MaskAugment(min_dimensions: ndarray | None = None, max_dimensions: ndarray | None = None, lower: ndarray | None = None, upper: ndarray | None = None, use_geo_boundaries: bool = False, center_mode: str = 'uniform', center_spread: ndarray | None = None, center_feature_index: int = 0)

Generic class to handle cutting out regions of an image.

Methods

__call__(data, meta, keys, context)	Apply an augmentation module.
apply(data, meta, keys, context)	Randomly mask a portion of the image.
cm_to_voxel(coords_cm, meta, dtype)	Convert detector coordinates at voxel centers back to indices.
generate_mask(data, meta, keys)	Generate a masking box metadata to apply to voxel index sets.
make_grid_aligned_meta(meta, lower_bound, ...)	Build a box metadata object snapped to the source voxel grid.
make_snapped_meta(meta, size, count, lower)	Build metadata snapped to the source grid from a proposed lower edge.
parse_optional_vector(value, name)	Parse an optional scalar-or-vector parameter into a length-3 array.
resolve_activity_center(data, keys, meta[, ...])	Estimate an activity center from all coordinate-carrying tensors.
resolve_activity_stats(data, keys, meta[, ...])	Estimate activity center and spread from coordinate-carrying tensors.
resolve_center(meta[, center, use_geo_center])	Resolve the pivot center for a geometric transform.
sample_box_lower(lower, upper, dimensions[, ...])	Sample the lower corner of a crop/mask box.
voxel_to_cm(coords, meta)	Convert voxel indices to detector coordinates at voxel centers.

name = 'mask'

apply(data: dict[str, Any], meta: ImageMeta3D, keys: list[str], context: dict[str, Any]) → tuple[dict[str, Any], ImageMeta3D]

Randomly mask a portion of the image.

Parameters:

• data (dict) – Dictionary of event data products to augment

• meta (Meta) – Shared image metadata

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• context (dict) – Shared augmentation context

Returns:

Updated data dictionary and unchanged metadata

Return type:

Tuple[Dict[str, Any], Meta]

generate_mask(data: dict[str, Any], meta: ImageMeta3D, keys: list[str]) → ImageMeta3D

Generate a masking box metadata to apply to voxel index sets.

Parameters:

• data (dict) – Dictionary of event data products used to estimate an activity center when activity-biased sampling is enabled

• meta (Meta) – Metadata of the original image

• keys (List[str]) – Keys corresponding to data products that carry coordinates

Returns:

Metadata describing the masked box

Return type:

Meta

class spine.io.augment.RotateAugment(axes: tuple[int, int] = (0, 1), k: int | None = None, center: ndarray | None = None, use_geo_center: bool = False, keep_meta: bool = True)

Generic class to handle right-angle image rotations.

Methods

__call__(data, meta, keys, context)	Apply an augmentation module.
apply(data, meta, keys, context)	Rotate the image by quarter turns in the requested plane.
apply_image_frame_rotation(data, meta, keys, k)	Apply the historical image-frame rotation behavior.
cm_to_voxel(coords_cm, meta, dtype)	Convert detector coordinates at voxel centers back to indices.
generate_centered_meta(meta, pivot, k)	Generate metadata for a rotation about an explicit pivot.
generate_meta(meta, k)	Generate the metadata for the rotated image.
make_grid_aligned_meta(meta, lower_bound, ...)	Build a box metadata object snapped to the source voxel grid.
make_snapped_meta(meta, size, count, lower)	Build metadata snapped to the source grid from a proposed lower edge.
parse_optional_vector(value, name)	Parse an optional scalar-or-vector parameter into a length-3 array.
resolve_activity_center(data, keys, meta[, ...])	Estimate an activity center from all coordinate-carrying tensors.
resolve_activity_stats(data, keys, meta[, ...])	Estimate activity center and spread from coordinate-carrying tensors.
resolve_center(meta[, center, use_geo_center])	Resolve the pivot center for a geometric transform.
rotate_coords(coords, count, k)	Rotate voxel coordinates by quarter turns.
rotate_points(points, pivot, k)	Rotate detector coordinates by quarter turns around a pivot.
sample_box_lower(lower, upper, dimensions[, ...])	Sample the lower corner of a crop/mask box.
sample_k()	Sample the number of quarter turns to apply.
voxel_to_cm(coords, meta)	Convert voxel indices to detector coordinates at voxel centers.

name = 'rotate'

apply(data: dict[str, Any], meta: ImageMeta3D, keys: list[str], context: dict[str, Any]) → tuple[dict[str, Any], ImageMeta3D]

Rotate the image by quarter turns in the requested plane.

Parameters:

• data (dict) – Dictionary of event data products to augment

• meta (Meta) – Shared image metadata

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• context (dict) – Shared augmentation context

Returns:

Updated data dictionary and rotated metadata

Return type:

Tuple[Dict[str, Any], Meta]

generate_centered_meta(meta: ImageMeta3D, pivot: ndarray, k: int) → ImageMeta3D

Generate metadata for a rotation about an explicit pivot.

Parameters:

• meta (Meta) – Metadata of the image before rotation

• pivot (np.ndarray) – (3,) Rotation center in detector coordinates (cm)

• k (int) – Number of 90-degree turns to apply

Returns:

Metadata of the rotated image volume

Return type:

Meta

apply_image_frame_rotation(data: dict[str, Any], meta: ImageMeta3D, keys: list[str], k: int) → tuple[dict[str, Any], ImageMeta3D]

Apply the historical image-frame rotation behavior.

Parameters:

• data (dict) – Dictionary of event data products to rotate

• meta (Meta) – Shared image metadata before rotation

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• k (int) – Number of 90-degree turns to apply

Returns:

Updated data dictionary and rotated metadata

Return type:

Tuple[Dict[str, Any], Meta]

sample_k() → int

Sample the number of quarter turns to apply.

Parameters:

None

Returns:

Number of 90-degree turns to apply

Return type:

int

rotate_coords(coords: ndarray, count: ndarray, k: int) → ndarray

Rotate voxel coordinates by quarter turns.

Parameters:

• coords (np.ndarray) – Voxel coordinates to rotate

• count (np.ndarray) – Original voxel counts along each axis

• k (int) – Number of 90-degree turns to apply

Returns:

Rotated voxel coordinates

Return type:

np.ndarray

rotate_points(points: ndarray, pivot: ndarray, k: int) → ndarray

Rotate detector coordinates by quarter turns around a pivot.

Parameters:

• points (np.ndarray) – (N, 3) Detector coordinates in cm

• pivot (np.ndarray) – (3,) Rotation center in detector coordinates (cm)

• k (int) – Number of 90-degree turns to apply

Returns:

(N, 3) Rotated detector coordinates in cm

Return type:

np.ndarray

generate_meta(meta: ImageMeta3D, k: int) → ImageMeta3D

Generate the metadata for the rotated image.

Parameters:

• meta (Meta) – Metadata of the image before rotation

• k (int) – Number of 90-degree turns to apply

Returns:

Metadata of the rotated image

Return type:

Meta

class spine.io.augment.TranslateAugment(lower: ndarray | None = None, upper: ndarray | None = None, use_geo: bool = False)

Generic class to handle moving images around.

Methods

__call__(data, meta, keys, context)	Apply an augmentation module.
apply(data, meta, keys, context)	Move an image around within the pre-defined volume.
cm_to_voxel(coords_cm, meta, dtype)	Convert detector coordinates at voxel centers back to indices.
generate_offset(meta, target_meta)	Generate a voxel offset within the target bounding box.
get_target_meta(meta[, original_meta])	Resolve the target translation volume metadata.
make_grid_aligned_meta(meta, lower_bound, ...)	Build a box metadata object snapped to the source voxel grid.
make_snapped_meta(meta, size, count, lower)	Build metadata snapped to the source grid from a proposed lower edge.
parse_optional_vector(value, name)	Parse an optional scalar-or-vector parameter into a length-3 array.
resolve_activity_center(data, keys, meta[, ...])	Estimate an activity center from all coordinate-carrying tensors.
resolve_activity_stats(data, keys, meta[, ...])	Estimate activity center and spread from coordinate-carrying tensors.
resolve_center(meta[, center, use_geo_center])	Resolve the pivot center for a geometric transform.
sample_box_lower(lower, upper, dimensions[, ...])	Sample the lower corner of a crop/mask box.
voxel_to_cm(coords, meta)	Convert voxel indices to detector coordinates at voxel centers.

name = 'translate'

apply(data: dict[str, Any], meta: ImageMeta3D, keys: list[str], context: dict[str, Any]) → tuple[dict[str, Any], ImageMeta3D]

Move an image around within the pre-defined volume.

Parameters:

• data (dict) – Dictionary of event data products to augment

• meta (Meta) – Shared image metadata

• keys (List[str]) – Keys corresponding to data products that carry coordinates

• context (dict) – Shared augmentation context

Returns:

Updated data dictionary and translated metadata

Return type:

Tuple[Dict[str, Any], Meta]

get_target_meta(meta: ImageMeta3D, original_meta: ImageMeta3D | None = None) → ImageMeta3D

Resolve the target translation volume metadata.

Parameters:

• meta (Meta) – Current image metadata

• original_meta (Meta, optional) – Original pre-augmentation metadata

Returns:

Metadata describing the translation target volume

Return type:

Meta

generate_offset(meta: ImageMeta3D, target_meta: ImageMeta3D) → ndarray

Generate a voxel offset within the target bounding box.

Parameters:

• meta (Meta) – Metadata of the image to translate

• target_meta (Meta) – Metadata of the translation target volume

Returns:

Integer voxel offset to apply along each axis

Return type:

np.ndarray