API¶

First, we provide a summary of the main functionalities of the package. Then we provide detailed documentation of every public function of ultrack.

Object Oriented API¶

ultrack.Tracker(config)

An Ultrack wrapper for its core functionalities.

Core functionalities¶

`ultrack.track`(config, *[, labels, sigma, ...])	All-in-one function for cell tracking, it accepts multiple inputs (labels or contours) and run all intermediate steps, computing segmentation hypothesis, linking and solving the ILP.
`ultrack.segment`(foreground, contours, config)	Add candidate segmentation (nodes) from foreground and edge to database.
`ultrack.link`(config[, images, scale, ...])	Links candidate segments (nodes) with their neighbors on the next time.
`ultrack.solve`(config[, batch_index, ...])	Compute tracking by selecting nodes with maximum flow from database.
`ultrack.add_flow`(config, vector_field)	Adds vector field (coordinate shift) data into nodes.
`ultrack.load_config`(path)	Creates MainConfig from TOML file.

Image processing utilities¶

`ultrack.imgproc.PlantSeg`(model_name[, ...])	A class for performing boundary detection using the Plant-Seg model.
`ultrack.imgproc.detect_foreground`(image[, ...])	Detect foreground using morphological reconstruction by dilation and thresholding.
`ultrack.imgproc.inverted_edt`(mask[, ...])	Computes Euclidean distance transform (EDT), inverts and normalizes it.
`ultrack.imgproc.normalize`(image, gamma[, ...])	Normalize image to between [0, 1] and applies a gamma transform (x ^ gamma).
`ultrack.imgproc.robust_invert`(image[, ...])	Inverts an image robustly by first smoothing it with a gaussian filter and then normalizing it to [0, 1].
`ultrack.imgproc.tracks_properties`(segments)	Calculate properties of tracked regions from segmentation data.
`ultrack.imgproc.Cellpose`(**kwargs)
`ultrack.imgproc.sam.MicroSAM`([model_type, ...])	Using a SAM model, generates masks for the entire image.
`ultrack.imgproc.register_timelapse`(timelapse, *)	Register a timelapse sequence using phase cross correlation.
`ultrack.imgproc.flow.timelapse_flow`(images)	Compute vector field from timelapse.
`ultrack.utils.labels_to_contours`(labels[, ...])	Converts and merges a sequence of labels into ultrack input format (foreground and contours)

Exporting¶

`ultrack.core.export.to_ctc`(output_dir, config)	Exports tracking results to cell-tracking challenge (http://celltrackingchallenge.net) format.
`ultrack.core.export.to_trackmate`(config[, ...])	Exports tracking results to TrackMate XML format.
`ultrack.core.export.to_tracks_layer`(config)	Exports solution from database to napari tracks layer format.
`ultrack.core.export.tracks_to_zarr`(config, ...)	Exports segmentations masks to zarr array, track_df assign the track_id to their respective segments.

Core functionalities¶

class ultrack.Tracker(config)¶

An Ultrack wrapper for its core functionalities.

Parameters:

config (MainConfig) – The configuration parameters.

Variables:

config (MainConfig) – The configuration parameters.
status (TrackerStatus) – The status of the tracking process.

Examples

>>> import ultrack
>>> from ultrack import MainConfig
>>> config = MainConfig()
>>> foreground = ...
>>> contours = ...
>>> vector_field = ...
>>> tracker = ultrack.Tracker(config)
>>> tracker.segment(foreground=foreground, contours=contours)
>>> tracker.add_flow(vector_field=vector_field)
>>> tracker.link()
>>> tracker.solve()

add_flow(vector_field)¶

Adds vector field (coordinate shift) data into nodes. If there are fewer vector fields than dimensions, the last dimensions from (z,y,x) have priority. For example, if 2 vector fields are provided for a 3D data, only (y, x) are updated. Vector field shape, except t, can be different from the original image. When this happens, the indexing is done mapping the position and rounding.

Parameters:

data_config (DataConfig) – Data configuration parameters.
vector_field (Sequence[ArrayLike]) – Vector field arrays. Each array per coordinate or a single (T, D, (Z), Y, X)-array.

add_nodes_prob(indices, probs)¶

Add nodes’ probabilities to the segmentation/tracking database.

Parameters:

config (MainConfig) – Main configuration parameters.
indices (ArrayLike) – Nodes’ indices database index.
probs (ArrayLike) – Nodes’ probabilities.

export_by_extension(include_parents=True, include_node_ids=True)¶

Exports solution from database to napari tracks layer format.

Parameters:

config (MainConfig) – Configuration parameters.
include_parents (bool) – Flag to include parents track id for each track id.
include_ids (bool) – Flag to include node ids for each unit.

Returns:

Tracks dataframe and an lineage graph, mapping node_id -> parent_id.

Return type:

Tuple[pd.DataFrame, Dict[int, List[int]]]

link(images=(), scale=None, batch_index=None, overwrite=False)¶

Links candidate segments (nodes) with their neighbors on the next time.

Parameters:

config (MainConfig) – Configuration parameters.
images (Sequence[ArrayLike]) – Optinal sequence of images for color space filtering.
scale (Sequence[float]) – Optional scaling for nodes’ distances.
batch_index (Optional[int], optional) – Batch index for processing a subset of nodes, by default everything is processed.
overwrite (bool) – Cleans up linking database content before processing.

segment(contours, config, max_segments_per_time=1000000, batch_index=None, overwrite=False, insertion_throttle_rate=50)¶

Add candidate segmentation (nodes) from foreground and edge to database.

Parameters:

foreground (ArrayLike) – Foreground probability array of shape (T, (Z), Y, X)
contours (ArrayLike) – Contours array of shape (T, (Z), Y, X)
config (MainConfig) – Configuration parameters.
max_segments_per_time (int) – Upper bound of segments per time point.
batch_index (Optional[int], optional) – Batch index for processing a subset of nodes, by default everything is processed.
overwrite (bool) – Cleans up segmentation, linking, and tracking database content before processing.
insertion_throttle_rate (int) – Throttling rate for insertion, by default 50. Only used with non-sqlite databases.

solve(batch_index=None, overwrite=False, use_annotations=False)¶

Compute tracking by selecting nodes with maximum flow from database.

Parameters:

config (MainConfig) – Configuration parameters.
batch_index (Optional[int], optional) – Batch index for processing a subset of nodes, by default everything is processed.
overwrite (bool, optional) – Resets existing solution before processing.
use_annotations (bool, optional) – Use annotations to fix ILP variables, by default False

to_ctc(config, margin=0, scale=None, first_frame=None, dilation_iters=0, stitch_tracks=False, overwrite=False)¶

Exports tracking results to cell-tracking challenge (http://celltrackingchallenge.net) format.

Parameters:

output_dir (Path) – Output directory to save segmentation masks and lineage graph
config (DataConfig) – Configuration parameters.
scale (Optional[Tuple[float]], optional) – Optional scaling of output segmentation masks, by default None
margin (int) – Margin used to filter out nodes and splitting their tracklets
first_frame (Optional[ArrayLike], optional) – Optional first frame detection mask to select a subset of tracks (e.g. Fluo-N3DL-DRO), by default None
dilation_iters (int) – Iterations of radius 1 morphological dilations on labels, applied after scaling, by default 0.
stitch_tracks (bool, optional) – Stitches (connects) incomplete tracks nearby tracks on subsequent time point, by default False
overwrite (bool, optional) – Flag to overwrite existing output_dir content, by default False

to_networkx(children_to_parent=False)¶

Convert napari tracks layer tracks dataframe to networkx directed graph. By default, the edges are the parent to child relationships.

Parameters:

config (MainConfig) – Configuration parameters.
children_to_parent (bool) – If set, edges encode child to parent relationships.

Returns:

Networkx graph.

Return type:

nx.DiGraph

to_pandas(include_parents=True, include_node_ids=True)¶

Exports solution from database to napari tracks layer format.

Parameters:

config (MainConfig) – Configuration parameters.
include_parents (bool) – Flag to include parents track id for each track id.
include_ids (bool) – Flag to include node ids for each unit.

Returns:

Tracks dataframe and an lineage graph, mapping node_id -> parent_id.

Return type:

Tuple[pd.DataFrame, Dict[int, List[int]]]

to_trackmate()¶

Convert a pandas DataFrame representation of Napari track layer to TrackMate XML format. <ImageData/> need to be set manually in the output XML.

Parameters:: tracks_df (pd.DataFrame) – A DataFrame with columns track_id, id, parent_id, t, z, y, x. Cells that belong to the same track have the same track_id.
Returns:: A string representation of the XML in the TrackMate format.
Return type:: str

Examples

>>> tracks_df = pd.DataFrame(
...     [[1,0,12.0,49.0,49.0,1000001,-1,-1],
...     [1,1,12.0,49.0,32.0,2000001,-1,1000001],
...     [2,1,12.0,49.0,66.0,2000002,-1,1000001]],
...     columns=['track_id','t','z','y','x','id','parent_track_id','parent_id']
... )
>>> print(tracks_df)
   track_id  t     z     y     x       id  parent_track_id  parent_id
0         1  0  12.0  49.0  49.0  1000001               -1         -1
1         1  1  12.0  49.0  32.0  2000001               -1    1000001
2         2  1  12.0  49.0  66.0  2000002               -1    1000001
>>> tracks_layer_to_trackmate(tracks_df)
<?xml version="1.0" ?>
<TrackMate version="7.11.1">
    <Model spatialunits="pixels" timeunits="frames">
        <AllTracks>
            <Track TRACK_ID="1" NUMBER_SPOTS="2" NUMBER_GAPS="0" TRACK_START="0" TRACK_STOP="1" name="Track_1">
                <Edge SPOT_SOURCE_ID="1000001" SPOT_TARGET_ID="2000001" EDGE_TIME="0.5"/>
            </Track>
            <Track TRACK_ID="2" NUMBER_SPOTS="1" NUMBER_GAPS="0" TRACK_START="1" TRACK_STOP="1" name="Track_2">
                <Edge SPOT_SOURCE_ID="1000001" SPOT_TARGET_ID="2000002" EDGE_TIME="0.5"/>
            </Track>
        </AllTracks>
        <FilteredTracks>
            <TrackID TRACK_ID="1"/>
            <TrackID TRACK_ID="2"/>
        </FilteredTracks>
        <AllSpots>
            <SpotsInFrame frame="0">
                <Spot ID="1000001" QUALITY="1.0" VISIBILITY="1" name="1000001" FRAME="0" RADIUS="5.0" POSITION_X="49.0" POSITION_Y="49.0" POSITION_Z="12.0"/>
            </SpotsInFrame>
            <SpotsInFrame frame="1">
                <Spot ID="2000001" QUALITY="1.0" VISIBILITY="1" name="2000001" FRAME="1" RADIUS="5.0" POSITION_X="32.0" POSITION_Y="49.0" POSITION_Z="12.0"/>
                <Spot ID="2000002" QUALITY="1.0" VISIBILITY="1" name="2000002" FRAME="1" RADIUS="5.0" POSITION_X="66.0" POSITION_Y="49.0" POSITION_Z="12.0"/>
            </SpotsInFrame>
        </AllSpots>
        <FeatureDeclarations>
            ...
        </FeatureDeclarations>
    </Model>
    <Settings>
        <InitialSpotFilter feature="QUALITY" value="0.0" isabove="true"/>
        <SpotFilterCollection/>
        <TrackFilterCollection/>
        <ImageData filename="None" folder="None" width="0" height="0" depth="0" nslices="1" nframes="2" pixelwidth="1.0" pixelheight="1.0" voxeldepth="1.0" timeinterval="1.0"/>
    </Settings>
</TrackMate>

to_tracks_layer(include_parents=True, include_node_ids=True)¶

Exports solution from database to napari tracks layer format.

Parameters:

config (MainConfig) – Configuration parameters.
include_parents (bool) – Flag to include parents track id for each track id.
include_ids (bool) – Flag to include node ids for each unit.

Returns:

Tracks dataframe and an lineage graph, mapping node_id -> parent_id.

Return type:

Tuple[pd.DataFrame, Dict[int, List[int]]]

to_zarr(tracks_df, store_or_path=None, chunks=None, overwrite=False)¶

Exports segmentations masks to zarr array, track_df assign the track_id to their respective segments. By changing the store this function can be used to write zarr arrays into disk.

Parameters:

config (MainConfig) – Configuration parameters.
tracks_df (pd.DataFrame) – Tracks dataframe, must have track_id column and be indexed by node id.
store_or_path (Union[None, Store, Path, str], optional) – Zarr storage or output path, if not provided zarr.TempStore is used.
chunks (Optional[Tuple[int]], optional) – Chunk size, if not provided it chunks time with 1 and the spatial dimensions as big as possible.
overwrite (bool, optional) – If True, overwrites existing zarr array.

Returns:

Output zarr array.

Return type:

zarr.Array

track(*, labels=None, sigma=None, foreground=None, contours=None, images=(), scale=None, vector_field=None, overwrite='none')¶

All-in-one function for cell tracking, it accepts multiple inputs (labels or contours) and run all intermediate steps, computing segmentation hypothesis, linking and solving the ILP. The results must be queried using the export function of preference.

Note: Either labels or foreground and contours can be used as input, but not both.

Parameters:

config (MainConfig) – Tracking configuration parameters.
labels (Optional[ArrayLike], optional) – Segmentation labels of shape (T, (Z), Y, X), by default None
sigma (Optional[Union[Sequence[float], float]], optional) – Edge smoothing parameter (gaussian blur) for labels to contours conversion, by default None
foreground (Optional[ArrayLike], optional) – Foreground probability array of shape (T, (Z), Y, X), by default None
contours (Optional[ArrayLike], optional) – Contours array of shape (T, (Z), Y, X), by default None
images (Sequence[ArrayLike]) – Optinal sequence of images (T, (Z), Y, X) for color space filtering.
scale (Sequence[float]) – Optional scaling for nodes’ distances.
vector_field (Union[ArrayLike, Sequence[ArrayLike]]) – Vector field arrays. Each array per coordinate or a single (T, D, (Z), Y, X)-array.
overwrite (Literal[``”all”, ``"links", "solutions", "none"], optional) – Clear the corresponding data from the database, by default nothing is overwritten with “none” When not “none”, only the cleared and subsequent parts of the pipeline is executed.

ultrack.add_flow(config, vector_field)¶

Parameters:

data_config (DataConfig) – Data configuration parameters.
vector_field (Sequence[ArrayLike]) – Vector field arrays. Each array per coordinate or a single (T, D, (Z), Y, X)-array.

ultrack.add_new_node(config, time, mask, bbox=None, index=None, include_overlaps=True)¶

Adds a new node to the database.

NOTE: this is not taking node shifts or image features (color) into account.

Parameters:

config (MainConfig) – Ultrack configuration parameters.
time (int) – Time point of the node.
mask (ArrayLike) – Binary mask of the node.
bbox (Optional[ArrayLike], optional) – Bounding box of the node, (min_0, min_1, …, max_0, max_1, …). When provided it assumes the mask is a crop of the original image, by default None
index (Optional[int], optional) – Node index, otherwise it is automatically generated, and returned.
include_overlaps (bool, optional) – Include overlaps in the database, by default True When False it will allow oclusions between new node and existing nodes.

Returns:

New node index.

Return type:

int

ultrack.export_tracks_by_extension(config, filename, overwrite=False)¶

Export tracks to a file given the file extension.

Supported file extensions are .xml, .csv, .zarr, .dot, and .json. - .xml exports to a TrackMate compatible XML file. - .csv exports to a CSV file. - .zarr exports the tracks to dense segments in a zarr array format. - .dot exports to a Graphviz DOT file. - .json exports to a networkx JSON file.

Parameters:

filename (str or Path) – The name of the file to save the tracks to.
config (MainConfig) – The configuration object.
overwrite (bool, optional) – Whether to overwrite the file if it already exists, by default False.

Array utilities¶

ultrack.utils.array.array_apply(*in_arrays, out_array, func, axis=0, **kwargs)¶

Apply a function over a given dimension of an array.

Parameters:

in_arrays (ArrayLike) – Arrays to apply function to.
out_array (ArrayLike) – Array to store result of function.
func (function) – Function to apply over time.
axis (Union[Tuple[int], int], optional) – Axis of data to apply func, by default 0.
args (tuple) – Positional arguments to pass to func.
**kwargs – Keyword arguments to pass to func.

ultrack.utils.array.assert_same_length(**kwargs)¶: Validates if key-word arguments have the same length.

ultrack.utils.array.check_array_chunk(array)¶: Checks if chunked array has chunk size of 1 on time dimension.

ultrack.utils.array.create_zarr(shape, dtype, store_or_path=None, overwrite=False, default_store_type=<class 'zarr.storage.TempStore'>, chunks=None, **kwargs)¶

Create a zarr array of zeros.

Parameters:

shape (Tuple[int, ]) – Shape of the array.
dtype (np.dtype) – Data type of the array.
store_or_path (Optional[Union[Path, str]], optional) – Path to store the array, if None a zarr.MemoryStore is used, by default None
overwrite (bool, optional) – Overwrite existing file, by default False
chunks (Optional[Tuple[int]], optional) – Chunk size, if not provided it chunks time with 1 and the spatial dimensions as big as possible.

Returns:

Zarr array of zeros.

Return type:

zarr.Array

ultrack.utils.array.large_chunk_size(shape, dtype, max_size=2147483647)¶

Computes a large chunk size for a given shape and dtype. Large chunks improves the performance on Elastic Storage Systems (ESS). Leading dimension (time) will always be chunked as 1.

Parameters:

shape (Tuple[int]) – Input data shape.
dtype (Union[str, np.dtype]) – Input data type.
max_size (int, optional) – Reference maximum size, by default 2147483647

Returns:

Suggested chunk size.

Return type:

Tuple[int]

ultrack.utils.array.validate_and_overwrite_path(path, overwrite, msg_type)¶: Validates and errors existance of path (or dir) and overwrites it if requested.

Image processing utilities¶

class ultrack.imgproc.PlantSeg(model_name, model_update=False, device=None, patch=(80, 160, 160), stride_ratio=0.75, batch_size=None, preprocess_sigma=None, postprocess_sigma=None, scale_factor=None)¶

A class for performing boundary detection using the Plant-Seg model. Plant-Seg documentation for more details, https://github.com/hci-unihd/plant-seg

Parameters:

model_name (str) – Name of the pre-trained segmentation model.
model_update (bool, optional) – Update the model if True. Default is False.
device ({str, torch.device}, optional) – Device for model execution. If None, the default device is used. Default is None.
patch (tuple[int], optional) – Patch size for model inference. Default is (80, 160, 160).
stride_ratio (float, optional) – Stride ratio for patch sampling. Default is 0.75.
batch_size (int, optional) – Batch size for inference. Default is None.
preprocess_sigma (float, optional) – Sigma value for Gaussian preprocessing filter. Default is None.
postprocess_sigma (float, optional) – Sigma value for Gaussian postprocessing filter. Default is None.
scale_factor (tuple[float], optional) – Scaling factors for input images. Default is None.

See also

ArrayPredictor: Class for making predictions using the PlantSeg model.

Examples

>>> seg_model = PlantSeg(model_name='generic_light_sheet_3D_unet', batch_size=4)
>>> segmentation_result = seg_model(image_data)

__call__(image, transpose=None)¶

Perform boundary detection using the PlantSeg model.

Parameters:

image (ArrayLike) – Input image data as a numpy, cupy, or Dask array.
transpose (tuple[int], optional) – Axes permutation for the input image. Permutation is applied after pre-processing and before inference and reverted after inference. Default is None.

Returns:

Segmentation boundary probability map as a numpy array.

Return type:

np.ndarray

ultrack.imgproc.detect_foreground(image, voxel_size=None, sigma=15.0, remove_hist_mode=False, min_foreground=0.0, channel_axis=None)¶

Detect foreground using morphological reconstruction by dilation and thresholding.

Parameters:

image (ArrayLike) – Input image.
voxel_size (ArrayLike) – Array of voxel size (z, y, x).
sigma (float, optional) – Sigma used to estimate background, it will be divided by voxel size, by default 15.0
remove_hist_mode (bool, optional) – Removes histogram mode before computing otsu threshold, useful when background regions are being detected.
min_foreground (float, optional) – Minimum value of foreground pixels after background subtraction and smoothing, by default 0.0
channel_axis (Optional[int], optional) – When provided it will be used to compute the foreground mask for each channel separately and merge them.

Returns:

Binary foreground mask.

Return type:

ArrayLike

ultrack.imgproc.inverted_edt(mask, voxel_size=None, axis=None)¶

Computes Euclidean distance transform (EDT), inverts and normalizes it.

Parameters:

mask (ArrayLike) – Cells’ foreground mask.
voxel_size (Optional[ArrayLike], optional) – Voxel size, by default None
axis (Optional[int], optional) – Axis to compute the EDT, by default None

Returns:

Inverted and normalized EDT.

Return type:

ArrayLike

ultrack.imgproc.normalize(image, gamma, lower_q=0.001, upper_q=0.9999)¶

Normalize image to between [0, 1] and applies a gamma transform (x ^ gamma).

Parameters:

image (ArrayLike) – Images as an T,Y,X,C array.
gamma (float) – Expoent of gamma transform.
lower_q (float, optional) – Lower quantile for normalization.
upper_q (float, optional) – Upper quantile for normalization.

Returns:

Normalized array.

Return type:

ArrayLike

ultrack.imgproc.register_timelapse(timelapse, *, store_or_path=None, overwrite=False, to_device=<function <lambda>>, reference_channel=None, overlap_ratio=0.25, normalization=None, padding=None, **kwargs)¶

Parameters:

timelapse (ArrayLike) – Input timelapse sequence, T(CZ)YX array C and Z are optional. NOTE: when provided, C must be the second dimension after T.
store_or_path (Union[Store, str, None], optional) – Zarr storage or a file path, to save the output, useful for larger than memory datasets. By default it loads the data into memory.
overwrite (bool, optional) – Overwrite output file if it already exists, when using directory store or a path.
to_device (Callable[[ArrayLike], ArrayLike], optional) – Function to move the input data to cuda device, by default lambda x: x (CPU).
reference_channel (Optional[int], optional) – Reference channel for registration, by default None. It must be provided when it contains a channel dimension.
overlap_ratio (float, optional) – Overlap ratio for phase cross correlation, by default 0.25.
normalization (Optional[str], optional) – Normalization method for phase cross correlation, by default None.
padding (Optional[int], optional) – Padding for registration, by default None.
**kwargs – Additional arguments for phase cross correlation. See skimage.registration phase_cross_correlation. # noqa: E501

Returns:

Registered timelapse sequence.

Return type:

zarr.Array

ultrack.imgproc.robust_invert(image, voxel_size=None, sigma=1.0, lower_quantile=None, upper_quantile=0.9999, channel_axis=None)¶

Inverts an image robustly by first smoothing it with a gaussian filter and then normalizing it to [0, 1].

Parameters:

image (ArrayLike) – Input image.
voxel_size (ArrayLike) – Array of voxel size (z, y, x).
sigma (float, optional) – Sigma used to smooth the image, by default 1.0.
lower_quantile (Optional[float], optional) – Lower quantile used to clip the intensities, minimum used when None.
upper_quantile (Optional[float], optional) – Upper quantile used to clip the intensities, maximum used when None.
channel_axis (Optional[int], optional) – When provided it will invert each channel separately and merge them.

Returns:

Inverted and normalized image.

Return type:

ArrayLike

ultrack.imgproc.tracks_properties(segments, tracks_df=None, image=None, properties=('label', 'num_pixels', 'area', 'intensity_sum', 'intensity_mean', 'intensity_std', 'intensity_min', 'intensity_max'), channel_axis=None, scale=None, n_workers=1)¶

Calculate properties of tracked regions from segmentation data.

Parameters:

segments (ArrayLike) – Array-like object containing the segmented regions for each time point. Time must be the first dimension.
tracks_df (Optional[pd.DataFrame], default None) – DataFrame containing tracking information for regions. When provided, measurements are merged into this data frame.
image (Optional[Union[List[ArrayLike], ArrayLike]], default None) – Array-like object containing the image data for each time point. If provided, intensity-based properties will be calculated; otherwise, only geometric properties will be calculated.
properties (Tuple[str, ], default (``’num_pixels’, ``'area', 'intensity_sum',) –
‘intensity_mean’, ‘intensity_std’, ‘intensity_min’,
’intensity_max’)

Tuple of property names to be calculated. Available options include:
- ’num_pixels’: Number of pixels in the region.
- ’area’: Area of the region.
- ’intensity_sum’: Sum of pixel intensities within the region.
- ’intensity_mean’: Mean pixel intensity within the region.
- ’intensity_std’: Standard deviation of pixel intensities within the region.
- ’intensity_min’: Minimum pixel intensity within the region.
- ’intensity_max’: Maximum pixel intensity within the region.
channel_axis (Optional[int], default None) – If the image parameter is provided and it contains multi-channel data, this parameter specifies the axis containing the channels. If None, the data is assumed to be single-channel.
scale (Optional[ArrayLike], default None) – Array-like object containing the scaling factors for each dimension of segments. Must include time dimension. Used for converting pixel measurements to physical units.
n_workers (int, default 1) – Number of workers to use for parallel processing.

Returns:

DataFrame containing the calculated properties for each region at each time point.

Return type:

pd.DataFrame

Notes

The function uses regionprops_table from skimage.measure to calculate properties.
If image is None, only geometric properties will be calculated.
If image is provided, intensity-based properties will be calculated, provided they are included in the properties parameter.

Examples

>>> # Example usage with image data
>>> image_data = ...  # Provide image data here
>>> segments_data = ...  # Provide segmentation data here
>>> result_df = tracks_properties(segments_data, image=image_data)

>>> # Example usage without image data
>>> segments_data = ...  # Provide segmentation data here
>>> result_df = tracks_properties(segments_data)

ultrack.utils.labels_to_contours(labels, sigma=None, foreground_store_or_path=None, contours_store_or_path=None, overwrite=False)¶

Converts and merges a sequence of labels into ultrack input format (foreground and contours)

Parameters:

labels (Union[ArrayLike, Sequence[ArrayLike]]) – List of labels with equal shape.
sigma (Optional[Union[Sequence[float], float]], optional) – Contours smoothing parameter (gaussian blur), contours aren’t smoothed when not provided.
foreground_store_or_path (str, zarr.storage.Store, optional) – Zarr storage, it can be used with zarr.NestedDirectoryStorage to save the output into disk. By default it loads the data into memory.
contours_store_or_path (str, zarr.storage.Store, optional) – Zarr storage, it can be used with zarr.NestedDirectoryStorage to save the output into disk. By default it loads the data into memory.
overwrite (bool, optional) – Overwrite output output files if they already exist, by default False.

Returns:

Combined foreground and edges arrays.

Return type:

Tuple[ArrayLike, ArrayLike]

class ultrack.imgproc.sam.MicroSAM(model_type='vit_h_lm', device=None, points_per_side=64, points_per_batch=128, pred_iou_thresh=0.88, stability_score_thresh=0.5, stability_score_offset=0.75, box_nms_thresh=0.5, crop_n_layers=0, crop_nms_thresh=0.7, crop_overlap_ratio=0.3413333333333333, crop_n_points_downscale_factor=1, point_grids=None, min_mask_region_area=0, max_mask_region_prop=0.1, mutex_watershed=False, tile_shape=(512, 512), halo_shape=(128, 128))¶

Using a SAM model, generates masks for the entire image. Generates a grid of point prompts over the image, then filters low quality and duplicate masks. The default settings are chosen for SAM with a ViT-H backbone.

Parameters:

device (Union[str, th.device, None] = None,) – The device to run the model on. If None, defaults to the default.
points_per_side (int or None, optional) – The number of points to be sampled along one side of the image. The total number of points is points_per_side**2. If None, ‘point_grids’ must provide explicit point sampling.
points_per_batch (int, optional) – Sets the number of points run simultaneously by the model. Higher numbers may be faster but use more GPU memory.
pred_iou_thresh (float, optional) – A filtering threshold in [0,1], using the model’s predicted mask quality.
stability_score_thresh (float, optional) – A filtering threshold in [0,1], using the stability of the mask under changes to the cutoff used to binarize the model’s mask predictions.
stability_score_offset (float, optional) – The amount to shift the cutoff when calculating the stability score.
box_nms_thresh (float, optional) – The box IoU cutoff used by non-maximal suppression to filter duplicate masks.
crop_n_layers (int, optional) – If >0, mask prediction will be run again on crops of the image. Sets the number of layers to run, where each layer has 2**i_layer number of image crops.
crop_nms_thresh (float, optional) – The box IoU cutoff used by non-maximal suppression to filter duplicate masks between different crops.
crop_overlap_ratio (float, optional) – Sets the degree to which crops overlap. In the first crop layer, crops will overlap by this fraction of the image length. Later layers with more crops scale down this overlap.
crop_n_points_downscale_factor (int, optional) – The number of points-per-side sampled in layer n is scaled down by crop_n_points_downscale_factor**n.
point_grids (list[np.ndarray] or None, optional) – A list over explicit grids of points used for sampling, normalized to [0,1]. The nth grid in the list is used in the nth crop layer. Exclusive with points_per_side.
min_mask_region_area (int, optional) – If >0, postprocessing will be applied to remove disconnected regions and holes in masks with an area smaller than min_mask_region_area. Requires opencv.
max_mask_region_prop (float, optional) – Regions larger than max_mask_region_prop * image area will be removed.
mutex_watershed (bool, optional) – If True, uses a mutex watershed algorithm to generate masks.
tile_shape (Tuple[int, int], optional) – The shape of the tiles used to precompute image embeddings.
halo_shape (Tuple[int, int], optional) – The shape of the halo (overlap) around each tile used to precompute image embeddings.

__call__(image)¶

Estimate contour of objects of an image.

Background regions not assigned to any label are set to -1.

Parameters:: image (ArrayLike) – The input image to be processed.
Returns:: The processed image with contours derived from the identified masks.
Return type:: np.ndarray

ultrack.imgproc.sam.set_peak_maxima_prompt(sam, sigma, min_distance, threshold_rel, **kwargs)¶

Configure a function to locate peak maxima in a given image using the MicroSAM framework. The function first applies a Gaussian blur to the image, then finds local maxima, and finally processes the image using the MicroSAM framework with the identified maxima.

Parameters:

sam (MicroSAM) – A MicroSAM instance used for image processing.
sigma (float) – Standard deviation for Gaussian filter applied to the image.
min_distance (int) – Minimum number of pixels separating peaks. Peaks closer than this are considered a single peak.
threshold_rel (float) – Minimum intensity difference between the peak and the pixels surrounding the peak for it to be considered a true peak. It is a relative threshold with respect to the maximum intensity in the image.
**kwargs – Additional keyword arguments to pass to the peak_local_max function.

Returns:

A function that, when given an image, will return an image processed by MicroSAM with contours derived from the identified peak maxima.

Return type:

Callable

Examples

>>> sam_instance = ...  # Initialize or provide a MicroSAM instance.
>>> peak_maxima_function = set_peak_maxima_prompt(
>>>     sam_instance, sigma=1, min_distance=10, threshold_rel=0.1,
>>> )
>>> processed_image = peak_maxima_function(input_image)

Flow¶

ultrack.imgproc.flow.add_flow(config, vector_field)¶

Parameters:

data_config (DataConfig) – Data configuration parameters.
vector_field (Sequence[ArrayLike]) – Vector field arrays. Each array per coordinate or a single (T, D, (Z), Y, X)-array.

ultrack.imgproc.flow.advenct_field(field, sources, shape=None, invert=True)¶

Advenct points from sources through the provided field. Shape indicates the original shape (space) and sources. Useful when field is down scaled from the original space.

Parameters:

field (ArrayLike) – Field array with shape T x D x (Z) x Y x X
sources (th.Tensor) – Array of sources N x D
shape (tuple[int, ]) – When provided scales field accordingly, D-dimensional tuple.
invert (bool) – When true flow is multiplied by -1, resulting in reversal of the flow.

Returns:

Trajectories of sources N x T x D

Return type:

th.Tensor

ultrack.imgproc.flow.advenct_field_from_labels(field, label, invert=True)¶

Advenct points from segmentation labels centroid.

Parameters:

field (ArrayLike) – Field array with shape T x D x (Z) x Y x X
label (ArrayLike) – Label image.
invert (bool) – When true flow is multiplied by -1, resulting in reversal of the flow.

Returns:

Trajectories of sources N x T x D

Return type:

ArrayLike

ultrack.imgproc.flow.advenct_from_quasi_random(field, img_shape, n_samples, invert=True, device=None)¶

Advenct points from quasi random uniform distribution.

Parameters:

field (ArrayLike) – Field array with shape T x D x (Z) x Y x X
img_shape (Tuple[int, ]) – Must be D-dimensional.
n_samples (int) – Number of samples.
invert (bool) – When true flow is multiplied by -1, resulting in reversal of the flow.
device (Optional[th.device]) – Torch device, by default uses last GPU if available or mps.

Returns:

Trajectories of sources N x T x D

Return type:

ArrayLike

ultrack.imgproc.flow.apply_field(field, image)¶

Transform image using vector field. Image will be scaled to the field size.

Parameters:

field (th.Tensor) – Vector field (D, z, y, x)
image (th.Tensor) – Original image used to compute the vector field.

Returns:

Transformed image (z, y, x)

Return type:

th.Tensor

ultrack.imgproc.flow.flow_field(source, target, im_factor=4, grid_factor=4, num_iterations=1000, lr=0.01, n_scales=3, init_grid=None)¶

Compute the flow vector field T that minimizes the mean squared error between T(source) and target.

Parameters:

source (torch.Tensor) – Source image (C, Z, Y, X).
target (torch.Tensor) – Target image (C, Z, Y, X).
im_factor (int, optional) – Image space down scaling factor, by default 4.
grid_factor (int, optional) – Grid space down scaling factor, by default 4. Grid dimensions will be divided by both im_factor and grid_factor.
num_iterations (int, optional) – Number of gradient descent iterations, by default 1000.
lr (float, optional) – Learning rate (gradient descent step), by default 1e-2
n_scales (int, optional) – Number of scales used for multi-scale optimization, by default 3.

Returns:

Vector field array with shape (D, (Z / factor), Y / factor, X / factor)

Return type:

torch.Tensor

ultrack.imgproc.flow.identity_grid(shape)¶

Grid equivalent to a identity vector field (no flow).

Parameters:: shape (tuple[int, ]) – Grid shape.
Returns:: Tensor of shape (Z, Y, X, D)
Return type:: th.Tensor

ultrack.imgproc.flow.timelapse_flow(images, store_or_path=None, chunks=None, channel_axis=None, im_factor=4, grid_factor=4, num_iterations=1000, lr=0.01, n_scales=3, device=None)¶

Compute vector field from timelapse.

Parameters:

images (ArrayLike) – Timelapse images shape as (T, …).
store_or_path (Union[None, Store, Path, str], optional) – Zarr storage or output path, if not provided zarr.TempStore is used.
chunks (Optional[Tuple[int]], optional) – Chunk size, if not provided it chunks time with 1 and the spatial dimensions as big as possible.
channel_axis (Optional[int], optional) – Channel axis EXCLUDING TIME (first axis), e.g (T, C, Y, X) would have channel_axis=0. If not provided assumes first axis after time.
im_factor (int, optional) – Image space down scaling factor, by default 4.
grid_factor (int, optional) – Grid space down scaling factor, by default 4. Grid dimensions will be divided by both im_factor and grid_factor.
num_iterations (int, optional) – Number of gradient descent iterations, by default 2000.
lr (float, optional) – Learning rate (gradient descent step), by default 1e-4
n_scales (int, optional) – Number of scales used for multi-scale optimization, by default 3.
device (Optional[th.device], optional) – Torch device, by default uses last GPU if available or mps.

Returns:

Vector field array with shape (T, D, (Z), Y, X).

Return type:

zarr.Array

ultrack.imgproc.flow.trajectories_to_tracks(trajectories)¶

Converts trajectories to napari tracks format.

Parameters:: trajectories (th.Tensor) – Input N x T x D trajectories.
Returns:: Napari tracks (N x T) x (2 + D) array.
Return type:: np.ndarray

Tracks utilities¶

ultrack.tracks.add_track_ids_to_tracks_df(df)¶

Adds track_id and parent_track_id columns to forest df. Each maximal path receveis a unique track_id.

Parameters:: df (pd.DataFrame) – Forest defined by the parent_id column and the dataframe indices.
Returns:: Inplace modified input dataframe with additional columns.
Return type:: pd.DataFrame

ultrack.tracks.close_tracks_gaps(tracks_df, max_gap, max_radius, spatial_columns=['z', 'y', 'x'], scale=None, segments=None, segments_store_or_path=None, overwrite=False)¶

Close gaps between tracklets in the given DataFrame.

Parameters:

tracks_df (pd.DataFrame) – The DataFrame containing the tracks information.
max_gap (int) – The maximum gap size to close.
max_radius (float) – The maximum distance between the end of one tracklet and the start of the next tracklet.
spatial_columns (List[str]) – The names of the columns containing the spatial information.
scale (Optional[ArrayLike]) – The scaling factors for the spatial columns.
segments (Optional[ArrayLike]) – When provided, the function will update the segments labels to match the tracks.
segments_store_or_path (Union[Store, Path, str, None]) – The store or path to save the updated segments, if not provided in memory store is used.
overwrite (bool) – If True, overwrites the segments store if it already exists.

Returns:

The DataFrame containing the tracks information with the gaps closed.

Return type:

Union[pd.DataFrame, Tuple[pd.DataFrame, ArrayLike]]

ultrack.tracks.filter_short_sibling_tracks(tracks_df, min_length, segments=None, segments_store_or_path=None, overwrite=False)¶

Filter short tracks created from fake division tracks shorter than min_length.

This function tranverse the graph bottom up and remove tracks that are shorter than min_length upon divions, merging the remaining sibling track with their parent.

If both are shorter than min_length, they are not removed.

Parameters:

tracks_df (pd.DataFrame) –

DataFrame containing track information with columns:
”track_id” : Unique identifier for each track. “parent_track_id” : Identifier of the parent track in the forest. (Other columns may be present in the DataFrame but are not used in this function.)
min_length (int) – Minimum track length, below this value the track is removed.
segments (Optional[ArrayLike]) – Segmentation array to update the tracks.
segments_store_or_path (Union[Store, Path, str, None]) – Store or path to save the new segments.
overwrite (bool) – If True, overwrite the existing segmentation array.

Returns:

If segments is None, returns the modified tracks dataframe. If segments is provided, returns the modified tracks dataframe and the updated segments.

Return type:

Union[pd.DataFrame, Tuple[pd.DataFrame, ArrayLike]]

ultrack.tracks.get_paths_to_roots(tracks_df, graph=None, *, node_index=None, track_index=None)¶

Returns paths from node_index or track_index to roots. If node_index and track_index are None, returns all paths to roots.

Parameters:

tracks_df (pd.DataFrame) –

DataFrame containing track information with columns:
”track_id” : Unique identifier for each track. “parent_track_id” : Identifier of the parent track in the forest. (Other columns may be present in the DataFrame but are not used in this function.)
graph (Optional[Dict[int, int]], optional) – Inverted forest graph, if not provided it will be computed from tracks_df.
node_index (Optional[int], optional) – Node (dataframe) index to compute path to roots.
track_index (Optional[int], optional) – Track index (track_id column value) to compute path to roots.

Returns:

DataFrame containing paths to roots.

Return type:

pd.DataFrame

ultrack.tracks.get_subgraph(tracks_df, track_ids)¶

Get a subgraph from a forest of tracks represented as a DataFrame.

Parameters:

tracks_df (pd.DataFrame) –

DataFrame containing track information with columns:
”track_id” : Unique identifier for each track. “parent_track_id” : Identifier of the parent track in the forest. (Other columns may be present in the DataFrame but are not used in this function.)
track_ids (ArrayLike) – An array-like object containing the track IDs for which to extract the subgraph.

Returns:

A DataFrame containing the subgraph of tracks corresponding to the input track IDs.

Return type:

pd.DataFrame

Examples

>>> subgraph_df = get_subgraph(tracks_df, [3, 7, 10])

Notes

The input DataFrame ‘tracks_df’ should have at least two columns: “track_id” and “parent_track_id”, where “track_id” represents the unique identifier for each track, and “parent_track_id” represents the identifier of the parent track in the forest.

ultrack.tracks.inv_tracks_df_forest(df)¶

Returns track_id and parent_track_id leaves-to-root inverted forest (set of trees) graph structure.

Example: forest[child_id] = parent_id

ultrack.tracks.left_first_search(track_id, graph)¶

Perform a left-first traversal on a binary tree represented as a graph and return a list of track IDs in the order they are visited during the traversal.

Parameters:

track_id (int) – The ID of the track to start the traversal from.
graph (Dict[int, List[int]]) – The graph representing the binary tree. It is a dictionary where the keys are track IDs, and the values are lists of two child track IDs. The binary tree must have exactly two children for each node.

Returns:

A list of track IDs visited during the left-first traversal of the binary tree, with track_id being the starting point.

Return type:

List[int]

Example

>>> graph = {1: [2, 3], 2: [4, 5], 3: [6, 7], 4: None, 5: None, 6: None, 7: None}
>>> result = _left_first_search(1, graph)
>>> print(result)
[4, 2, 5, 1, 6, 3, 7]

ultrack.tracks.sort_track_ids(tracks_df)¶

Sort track IDs in a given DataFrame representing tracks in a way that maintains the left-first order of the binary tree formed by their parent-child relationships.

Parameters:: tracks_df (pd.DataFrame) – A DataFrame containing information about tracks, where each row represents a track and contains at least two columns - “track_id” and “track_parent_id”. The “track_id” column holds unique track IDs, and the “track_parent_id” column contains the parent track IDs for each track. The DataFrame should have a consistent parent-child relationship, forming one or multiple binary trees.
Returns:: A NumPy array containing the sorted track IDs based on the left-first traversal of the binary trees formed by the parent-child relationships.
Return type:: np.ndarray

Example

>>> import pandas as pd
>>> import numpy as np
>>> data = {
...     "track_id": [1, 2, 3, 4, 5, 6, 7],
...     "track_parent_id": [None, 1, 1, 2, 2, 3, 3],
... }
>>> tracks_df = pd.DataFrame(data)
>>> sorted_track_ids = sort_track_ids(tracks_df)
>>> print(sorted_track_ids)
[4 2 5 1 6 3 7]

ultrack.tracks.sort_trees_by_length(df, graph=None)¶

Sorts trees from the track graph by length (deepest tree path).

Parameters:

df (pd.DataFrame) – tracks dataframe.
graph (Dict[int, int], optional) – Child -> parent tracks graph. Optional, if not provided it will be computed from the dataframe, must have track_id and parent_track_id columns.

Returns:

Sorted list of tracks dataframe.

Return type:

List[pd.DataFrame]

ultrack.tracks.sort_trees_by_max_radius(df, scale=None, metric='euclidean')¶

Sorts trees from the track graph by radius (distance between nodes at the same time point).

Parameters:

df (pd.DataFrame) – tracks dataframe.
scale (ArrayLike, optional) – Spatial domain scale.
metric (Union[Callable, str], optional) – Distance metric, see scipy.spatial.distance.pdist for more information.

Returns:

Sorted list of tracks dataframe.

Return type:

List[pd.DataFrame]

ultrack.tracks.split_tracks_df_by_lineage(tracks_df)¶

Split tracks dataframe into a list of dataframes, one for each lineage, sorted by the root track id.

Parameters:

tracks_df (pd.DataFrame) –

Tracks dataframe with columns:: ”track_id” : Unique identifier for each track. “parent_track_id” : Identifier of the parent track in the forest. (Other columns may be present in the DataFrame but are not used in this function.)

Returns:

List of dataframes, one for each lineage.

Return type:

List[pd.DataFrame]

ultrack.tracks.split_trees(tracks_df)¶

Split tracks forest into trees.

Parameters:

tracks_df (pd.DataFrame) –

DataFrame containing track information with columns:: ”track_id” : Unique identifier for each track. “parent_track_id” : Identifier of the parent track in the forest. (Other columns may be present in the DataFrame but are not used in this function.)

Returns:

List of dataframes, each representing a tree.

Return type:

List[pd.DataFrame]

ultrack.tracks.tracks_df_forest(df, remove_roots=False, numba_dict=False)¶

Creates the forest graph of track lineages

Example: forest[parent_id] = [child_id_0, child_id_1]

Parameters:

df (pd.DataFrame) – Tracks dataframe.
remove_roots (bool) – If True, removes root nodes (nodes with no parent).
numba_dict (bool) – If True, returns a numba typed dictionary.

Returns:

Forest graph where parent maps to their children (parent -> children)

Return type:

Dict[int, List[int]]

ultrack.tracks.tracks_df_movement(tracks_df, lag=1, cols=None)¶

Compute the displacement for track data across given time lags.

This function computes the displacement (difference) for track coordinates across the specified lag periods.

NOTE: this sort the dataframe by [“track_id”, “t”].

Parameters:

tracks_df (pd.DataFrame) – Dataframe containing track data. It is expected to have columns [“track_id”, “t”] and any of [“z”, “y”, “x”] representing the 3D coordinates.
lag (int, optional) – Number of periods to compute the difference over. Default is 1.
cols (tuple[str, ], optional) – Columns to compute the displacement for. If not provided, it will try to find any of [“z”, “y”, “x”] columns in the dataframe and use them.

Returns:

Dataframe of the displacement (difference) of coordinates for the given lag. Displacements for the first row of each track_id will be set to zero.

Return type:

pd.DataFrame

Examples

>>> df = pd.DataFrame({
...     "track_id": [1, 1, 2, 2],
...     "t": [1, 2, 1, 2],
...     "z": [0, 1, 0, 2],
...     "y": [1, 2, 1, 2],
...     "x": [2, 3, 2, 2]
... })
>>> print(tracks_df_movement(df))
   z    y    x
0 0.0  0.0  0.0
1 1.0  1.0  1.0
2 0.0  0.0  0.0
3 2.0  1.0  0.0

ultrack.tracks.tracks_length(tracks_df, include_appearing=True, include_disappearing=True)¶

Compute the length of each track in a tracks dataframe.

Parameters:

tracks_df (pd.DataFrame) –

DataFrame containing track information with columns:
”t” : Time step index for each data point in the track. “track_id” : Unique identifier for each track. “parent_track_id” : Unique identifier for the parent track.
include_appearing (bool, optional) – Include tracks that appear outside the first time step, by default True.
include_disappearing (bool, optional) – Include tracks that disappear outside the last time step, by default True.

Returns:

Series containing the length of each track.

Return type:

pd.DataFrame

ultrack.tracks.tracks_profile_matrix(tracks_df, columns)¶

Construct a profile matrix from a pandas DataFrame containing tracks data.

Parameters:

tracks_df (pd.DataFrame) –

DataFrame containing track information with columns:
”track_id” : Unique identifier for each track. “t” : Time step index for each data point in the track. Other columns specified in ‘columns’ parameter, representing track attributes.
columns (List[str]) – List of strings, specifying the columns of ‘tracks_df’ to use as attributes.

Returns:

A 3D NumPy array representing the profile matrix with shape (num_attributes, num_tracks, max_timesteps), where ‘num_attributes’ is the number of attributes specified in ‘columns’, ‘num_tracks’ is the number of unique tracks, and ‘max_timesteps’ is the maximum number of timesteps encountered among all tracks.

Return type:

np.ndarray

Exporting¶

ultrack.core.export.to_ctc(output_dir, config, margin=0, scale=None, first_frame=None, dilation_iters=0, stitch_tracks=False, overwrite=False)¶

Exports tracking results to cell-tracking challenge (http://celltrackingchallenge.net) format.

Parameters:

output_dir (Path) – Output directory to save segmentation masks and lineage graph
config (DataConfig) – Configuration parameters.
scale (Optional[Tuple[float]], optional) – Optional scaling of output segmentation masks, by default None
margin (int) – Margin used to filter out nodes and splitting their tracklets
first_frame (Optional[ArrayLike], optional) – Optional first frame detection mask to select a subset of tracks (e.g. Fluo-N3DL-DRO), by default None
dilation_iters (int) – Iterations of radius 1 morphological dilations on labels, applied after scaling, by default 0.
stitch_tracks (bool, optional) – Stitches (connects) incomplete tracks nearby tracks on subsequent time point, by default False
overwrite (bool, optional) – Flag to overwrite existing output_dir content, by default False

ultrack.core.export.to_networkx(config, children_to_parent=False)¶

Convert solution from database to networkx directed graph. By default, the edges are the parent to child relationships.

Parameters:

config (MainConfig) – Configuration parameters.
children_to_parent (bool) – If set, edges encode child to parent relationships.

Returns:

Networkx graph.

Return type:

nx.DiGraph

ultrack.core.export.to_trackmate(config, output_path=None, overwrite=False)¶

Exports tracking results to TrackMate XML format.

Parameters:

config (MainConfig) – ULTrack configuration parameters.
output_path (Optional[Path], optional) – Output file path, by default None
overwrite (bool, optional) – Whether to overwrite the output file if it already exists, by default False

Returns:

A string representation of the XML in the TrackMate format.

Return type:

str

ultrack.core.export.to_tracks_layer(config, include_parents=True, include_node_ids=True)¶

Exports solution from database to napari tracks layer format.

Parameters:

config (MainConfig) – Configuration parameters.
include_parents (bool) – Flag to include parents track id for each track id.
include_ids (bool) – Flag to include node ids for each unit.

Returns:

Tracks dataframe and an lineage graph, mapping node_id -> parent_id.

Return type:

Tuple[pd.DataFrame, Dict[int, List[int]]]

ultrack.core.export.tracks_layer_to_networkx(tracks_df, children_to_parent=False)¶

Convert napari tracks layer tracks dataframe to networkx directed graph. By default, the edges are the parent to child relationships.

Parameters:

config (MainConfig) – Configuration parameters.
children_to_parent (bool) – If set, edges encode child to parent relationships.

Returns:

Networkx graph.

Return type:

nx.DiGraph

ultrack.core.export.tracks_layer_to_trackmate(tracks_df)¶

Convert a pandas DataFrame representation of Napari track layer to TrackMate XML format. <ImageData/> need to be set manually in the output XML.

Parameters:: tracks_df (pd.DataFrame) – A DataFrame with columns track_id, id, parent_id, t, z, y, x. Cells that belong to the same track have the same track_id.
Returns:: A string representation of the XML in the TrackMate format.
Return type:: str

Examples

>>> tracks_df = pd.DataFrame(
...     [[1,0,12.0,49.0,49.0,1000001,-1,-1],
...     [1,1,12.0,49.0,32.0,2000001,-1,1000001],
...     [2,1,12.0,49.0,66.0,2000002,-1,1000001]],
...     columns=['track_id','t','z','y','x','id','parent_track_id','parent_id']
... )
>>> print(tracks_df)
   track_id  t     z     y     x       id  parent_track_id  parent_id
0         1  0  12.0  49.0  49.0  1000001               -1         -1
1         1  1  12.0  49.0  32.0  2000001               -1    1000001
2         2  1  12.0  49.0  66.0  2000002               -1    1000001
>>> tracks_layer_to_trackmate(tracks_df)
<?xml version="1.0" ?>
<TrackMate version="7.11.1">
    <Model spatialunits="pixels" timeunits="frames">
        <AllTracks>
            <Track TRACK_ID="1" NUMBER_SPOTS="2" NUMBER_GAPS="0" TRACK_START="0" TRACK_STOP="1" name="Track_1">
                <Edge SPOT_SOURCE_ID="1000001" SPOT_TARGET_ID="2000001" EDGE_TIME="0.5"/>
            </Track>
            <Track TRACK_ID="2" NUMBER_SPOTS="1" NUMBER_GAPS="0" TRACK_START="1" TRACK_STOP="1" name="Track_2">
                <Edge SPOT_SOURCE_ID="1000001" SPOT_TARGET_ID="2000002" EDGE_TIME="0.5"/>
            </Track>
        </AllTracks>
        <FilteredTracks>
            <TrackID TRACK_ID="1"/>
            <TrackID TRACK_ID="2"/>
        </FilteredTracks>
        <AllSpots>
            <SpotsInFrame frame="0">
                <Spot ID="1000001" QUALITY="1.0" VISIBILITY="1" name="1000001" FRAME="0" RADIUS="5.0" POSITION_X="49.0" POSITION_Y="49.0" POSITION_Z="12.0"/>
            </SpotsInFrame>
            <SpotsInFrame frame="1">
                <Spot ID="2000001" QUALITY="1.0" VISIBILITY="1" name="2000001" FRAME="1" RADIUS="5.0" POSITION_X="32.0" POSITION_Y="49.0" POSITION_Z="12.0"/>
                <Spot ID="2000002" QUALITY="1.0" VISIBILITY="1" name="2000002" FRAME="1" RADIUS="5.0" POSITION_X="66.0" POSITION_Y="49.0" POSITION_Z="12.0"/>
            </SpotsInFrame>
        </AllSpots>
        <FeatureDeclarations>
            ...
        </FeatureDeclarations>
    </Model>
    <Settings>
        <InitialSpotFilter feature="QUALITY" value="0.0" isabove="true"/>
        <SpotFilterCollection/>
        <TrackFilterCollection/>
        <ImageData filename="None" folder="None" width="0" height="0" depth="0" nslices="1" nframes="2" pixelwidth="1.0" pixelheight="1.0" voxeldepth="1.0" timeinterval="1.0"/>
    </Settings>
</TrackMate>

ultrack.core.export.tracks_to_zarr(config, tracks_df, store_or_path=None, chunks=None, overwrite=False)¶

Exports segmentations masks to zarr array, track_df assign the track_id to their respective segments. By changing the store this function can be used to write zarr arrays into disk.

Parameters:

config (MainConfig) – Configuration parameters.
tracks_df (pd.DataFrame) – Tracks dataframe, must have track_id column and be indexed by node id.
store_or_path (Union[None, Store, Path, str], optional) – Zarr storage or output path, if not provided zarr.TempStore is used.
chunks (Optional[Tuple[int]], optional) – Chunk size, if not provided it chunks time with 1 and the spatial dimensions as big as possible.
overwrite (bool, optional) – If True, overwrites existing zarr array.

Returns:

Output zarr array.

Return type:

zarr.Array