piscis.core

Classes

_Piscis	Base class for running the Piscis algorithm.
Piscis	Class for running the Piscis algorithm.
PiscisLegacy	Class for running the PyTorch port of the legacy Piscis algorithm originally implemented in JAX.

Functions

adjust_parameters(→ numpy.ndarray)

Adjust tunable parameters for a given set of intermediate feature maps.

Module Contents

class piscis.core._Piscis(model_name: str, batch_size: int, input_size: Tuple[int, int] | None, device: str | torch.device | None, pooling: str)

Base class for running the Piscis algorithm.

Attributes:

model_namestr

Model name.

batch_sizeint

Batch size for the CNN.

adjustmentstr

Adjustment type applied to images during preprocessing.

input_sizeTuple[int, int]

Input size for the CNN.

dilation_iterationsint

Number of iterations used to dilate ground truth labels during training.

channelsint

Number of channels in the input images.

deviceOptional[Union[str, torch.device]]

Device to run the model on.

poolingstr

Pooling type applied to labels.

modelSpotsModel

Model.

model_name

batch_size

device

pooling

model

_predict(x: numpy.ndarray | dask.array.Array, stack: bool, scale: float, threshold: float, min_distance: int, intermediates: bool) → Tuple[numpy.ndarray, xarray.DataArray] | numpy.ndarray

Predict spots in an image or stack of images.

Parameters:

xnp.ndarray or da.Array

Image or stack of images.

stackbool, optional

Whether x is a stack of images.

scalefloat, optional

Scale factor for rescaling x.

thresholdfloat, optional

Spot detection threshold.

min_distanceint, optional

Minimum distance between spots.

intermediatesbool, optional

Whether to return intermediate feature maps.

Returns:

coordsnp.ndarray

Predicted spot coordinates.

ynp.ndarray, optional

Intermediate feature maps. Only returned if intermediates is True.

_predict_stack(tiles: deeptile.core.data.Tiled, stack_axis_len: int, scales: numpy.ndarray, threshold: float, min_distance: int, intermediates: bool) → Tuple[deeptile.Output, numpy.ndarray] | deeptile.Output

Predict spots in a stack of images.

Parameters:

tilesTiled

Tiles of images.

stack_axis_lenint

Length of the stack axis.

scalesnp.ndarray

Scales for rescaling tiles.

thresholdfloat

Spot detection threshold. Can be interpreted as the minimum number of fully confident pixels necessary to identify a spot.

min_distanceint

Minimum distance between spots.

intermediatesbool

Whether to return intermediate feature maps.

Returns:

coordsOutput

Predicted spot coordinates.

ynp.ndarray, optional

Intermediate feature maps. Only returned if intermediates is True.

_predict_plane(tiles: deeptile.core.data.Tiled, scales: numpy.ndarray, threshold: float, min_distance: int, intermediates: bool) → Tuple[deeptile.Output, numpy.ndarray] | deeptile.Output

Predict spots in a single plane.

Parameters:

tilesTiled

Tiles of images.

scalesnp.ndarray

Scales for rescaling tiles.

thresholdfloat

Spot detection threshold. Can be interpreted as the minimum number of fully confident pixels necessary to identify a spot.

min_distanceint

Minimum distance between spots.

intermediatesbool

Whether to return intermediate feature maps.

Returns:

coordsOutput

Predicted spot coordinates.

ynp.ndarray, optional

Intermediate feature maps. Only returned if intermediates is True.

_preprocess(x: numpy.ndarray | dask.array.Array, stack: bool) → Tuple[dask.array.Array, bool]

Preprocess the input.

Parameters:

xnp.ndarray or dask.array.Array

Image or stack of images.

stackbool

Whether x is a stack of images.

Returns:

xUnion[np.ndarray, da.Array]

Preprocessed image or stack of images.

batch_axisbool

Whether x has a batch axis.

Raises:

ValueError

If x does not have the correct dimensions.

ValueError

If x does not have the correct number of channels.

_process(tiles: deeptile.core.data.Tiled) → Tuple[numpy.ndarray, numpy.ndarray]

Process tiles.

Parameters:

tilesTiled

Tiles of images.

Returns:

labelsnp.ndarray

Predicted labels.

deltasnp.ndarray

Predicted displacement vectors.

static _postprocess_stack(labels: numpy.ndarray, deltas: numpy.ndarray, scales: numpy.ndarray, threshold: float, min_distance: int) → deeptile.Output

Postprocess tiles.

Parameters:

labelsnp.ndarray

Predicted labels.

deltasnp.ndarray

Predicted displacement vectors.

scalesnp.ndarray

Scales for rescaling tiles.

thresholdfloat

Spot detection threshold. Can be interpreted as the minimum number of fully confident pixels necessary to identify a spot.

min_distanceint

Minimum distance between spots.

Returns:

coordsOutput

Predicted spot coordinates.

_apply(x: torch.Tensor) → Tuple[torch.Tensor, torch.Tensor]

Apply SpotsModel to a batch images.

xtorch.Tensor

Batch of images.

Returns:

labelsnp.ndarray

Predicted binary labels.

deltasnp.ndarray

Predicted displacements vectors.

class piscis.core.Piscis(model_name: str = '20251212', batch_size: int = 1, input_size: Tuple[int, int] | None = None, device: str | torch.device | None = None)

Bases: _Piscis

Class for running the Piscis algorithm.

Attributes:

model_namestr

Model name.

batch_sizeint

Batch size for the CNN.

adjustmentstr

Adjustment type applied to images during preprocessing.

input_sizeTuple[int, int]

Input size for the CNN.

dilation_iterationsint

Number of iterations used to dilate ground truth labels during training.

channelsint

Number of channels in the input images.

deviceOptional[Union[str, torch.device]]

Device to run the model on.

poolingstr

Pooling type applied to labels.

modelSpotsModel

Model.

predict(x: numpy.ndarray | dask.array.Array, stack: bool = False, scale: float = 1.0, threshold: float = 0.5, min_distance: int = 1, intermediates: bool = False) → Tuple[numpy.ndarray, xarray.DataArray] | numpy.ndarray

Predict spots in an image or stack of images.

Parameters:

xnp.ndarray or da.Array

Image or stack of images.

stackbool, optional

Whether x is a stack of images. Default is False.

scalefloat, optional

Scale factor for rescaling x. Default is 1.

thresholdfloat, optional

Spot detection threshold. Default is 0.5.

min_distanceint, optional

Minimum distance between spots. Default is 1.

intermediatesbool, optional

Whether to return intermediate feature maps. Default is False.

Returns:

coordsnp.ndarray

Predicted spot coordinates.

ynp.ndarray, optional

Intermediate feature maps. Only returned if intermediates is True.

class piscis.core.PiscisLegacy(model_name: str = '20230905', batch_size: int = 1, input_size: Tuple[int, int] | None = None, device: str | torch.device | None = None)

Bases: _Piscis

Class for running the PyTorch port of the legacy Piscis algorithm originally implemented in JAX.

Attributes:

model_namestr

Model name.

batch_sizeint

Batch size for the CNN.

adjustmentstr

Adjustment type applied to images during preprocessing.

input_sizeTuple[int, int]

Input size for the CNN.

dilation_iterationsint

Number of iterations used to dilate ground truth labels during training.

channelsint

Number of channels in the input images.

deviceOptional[Union[str, torch.device]]

Device to run the model on.

poolingstr

Pooling type applied to labels.

modelSpotsModel

Model.

predict(x: numpy.ndarray | dask.array.Array, stack: bool = False, scale: float = 1.0, threshold: float = 1.0, min_distance: int = 1, intermediates: bool = False) → Tuple[numpy.ndarray, xarray.DataArray] | numpy.ndarray

Predict spots in an image or stack of images.

Parameters:

xnp.ndarray or da.Array

Image or stack of images.

stackbool, optional

Whether x is a stack of images. Default is False.

scalefloat, optional

Scale factor for rescaling x. Default is 1.

thresholdfloat, optional

Spot detection threshold. Can be interpreted as the minimum number of fully confident pixels necessary to identify a spot. Default is 1.0.

min_distanceint, optional

Minimum distance between spots. Default is 1.

intermediatesbool, optional

Whether to return intermediate feature maps. Default is False.

Returns:

coordsnp.ndarray

Predicted spot coordinates.

ynp.ndarray, optional

Intermediate feature maps. Only returned if intermediates is True.

piscis.core.adjust_parameters(y: xarray.DataArray, threshold: float = 0.5, min_distance: int = 1) → numpy.ndarray

Adjust tunable parameters for a given set of intermediate feature maps.

Parameters:

yxr.DataArray

Intermediate feature maps.

threshold: float

Spot detection threshold. Default is 0.5.

min_distanceint, optional

Minimum distance between spots. Default is 1.

Returns:

coordsnp.ndarray

Predicted spot coordinates.