backend.solve
Attributes
Functions
|
Get a tracking solution for the given segmentation and parameters. |
|
Build the candidate graph from input data. |
|
Solve the tracking problem on the full candidate graph at once. |
|
Solve a single window subgraph. |
|
Solve a single window for interactive parameter testing. |
|
Solve the tracking problem in chunks using a sliding window approach. |
|
Set PIN_ATTR on candidate graph nodes/edges in the overlap region. |
|
Construct a motile solver with the parameters specified in the solver |
|
Return the name of the ILP solver backend that will be used. |
Module Contents
- backend.solve.logger
- backend.solve.PIN_ATTR = 'pinned'
- backend.solve.solve(solver_params: backend.solver_params.SolverParams, input_data: numpy.ndarray, on_solver_update: collections.abc.Callable | None = None, scale: list | None = None, cand_graph: tracksdata.graph.GraphView | None = None) tracksdata.graph.GraphView
Get a tracking solution for the given segmentation and parameters.
Constructs a candidate graph from the segmentation (unless one is provided), a solver from the parameters, and then runs solving and returns a networkx graph with the solution. Most of this functionality is implemented in the motile toolbox.
- Args:
- solver_params (SolverParams): The solver parameters to use when
initializing the solver
- input_data (np.ndarray): The input segmentation or points list to run
tracking on. If 2D, assumed to be a list of points, otherwise a segmentation.
- on_solver_update (Callable, optional): A function that is called
whenever the motile solver emits an event. The function should take a dictionary of event data, and can be used to track progress of the solver. Defaults to None.
scale (list, optional): The scale of the data in each dimension. cand_graph (td.graph.GraphView, optional): A pre-built candidate graph. If
provided, skips candidate graph construction (except for single-window mode which always builds its own). Defaults to None.
- Returns:
- td.graph.GraphView: A solution graph where the ids of the nodes correspond to
the time and ids of the passed in segmentation labels. See funtracks for exact implementation details.
- backend.solve.build_candidate_graph(input_data: numpy.ndarray, solver_params: backend.solver_params.SolverParams, scale: list | None = None) tracksdata.graph.GraphView
Build the candidate graph from input data.
- backend.solve._solve_full(cand_graph: tracksdata.graph.GraphView, solver_params: backend.solver_params.SolverParams, on_solver_update: collections.abc.Callable | None = None) tracksdata.graph.GraphView
Solve the tracking problem on the full candidate graph at once.
- backend.solve._solve_window(window_subgraph: tracksdata.graph.GraphView, solver_params: backend.solver_params.SolverParams, on_solver_update: collections.abc.Callable | None = None) tracksdata.graph.GraphView | None
Solve a single window subgraph.
This is the core solving logic shared by both single window mode and chunked solving.
- Args:
- window_subgraph: The subgraph for this window. If any nodes or edges
have the PIN_ATTR attribute set, a Pin constraint will be used.
solver_params: The solver parameters. on_solver_update: Callback for solver progress updates.
- Returns:
The solution graph for this window, or None if the window has no nodes.
- backend.solve._solve_single_window(input_data: numpy.ndarray, solver_params: backend.solver_params.SolverParams, on_solver_update: collections.abc.Callable | None = None, scale: list | None = None) tracksdata.graph.GraphView
Solve a single window for interactive parameter testing.
Builds the full candidate graph, filters it to the window frames, and solves. Node times are naturally correct (no adjustment needed).
- Args:
input_data: The full input segmentation or points list. solver_params: The solver parameters including window_size and single_window_start. on_solver_update: Callback for solver progress updates. scale: The scale of the data in each dimension.
- Returns:
The solution graph for the requested window.
- Raises:
ValueError: If single_window_start is beyond the data range.
- backend.solve._solve_chunked(cand_graph: tracksdata.graph.GraphView, solver_params: backend.solver_params.SolverParams, on_solver_update: collections.abc.Callable | None = None) tracksdata.graph.GraphView
Solve the tracking problem in chunks using a sliding window approach.
This function solves the tracking problem in windows of window_size frames, with overlap_size frames of overlap between consecutive windows. The overlap region from the previous window is pinned (fixed) when solving the next window to maintain consistency across windows.
- Args:
cand_graph: The full candidate graph with all nodes and edges. solver_params: The solver parameters including window_size and overlap_size. on_solver_update: Callback for solver progress updates.
- Returns:
The combined solution graph from all windows.
- backend.solve._set_pinning_on_graph(cand_graph: tracksdata.graph.GraphView, solution_graph: tracksdata.graph.GraphView, overlap_start: int, overlap_end: int) None
Set PIN_ATTR on candidate graph nodes/edges in the overlap region.
For all nodes and edges in the overlap region [overlap_start, overlap_end), sets PIN_ATTR to True if selected in the solution, False if not selected.
- Args:
cand_graph: The full candidate graph to modify in place. solution_graph: The solution graph from the current window. overlap_start: Start frame of overlap region (inclusive). overlap_end: End frame of overlap region (exclusive).
- backend.solve._SKIP_ATTRS
- backend.solve.construct_solver(cand_graph: tracksdata.graph.GraphView, solver_params: backend.solver_params.SolverParams) motile.Solver
Construct a motile solver with the parameters specified in the solver params object.
- Args:
cand_graph (td.graph.GraphView): The candidate graph to use in the solver solver_params (SolverParams): The costs and constraints to use in
the solver
- Returns:
- Solver: A motile solver with the specified graph, costs, and
constraints.
- backend.solve.get_solver_name() str
Return the name of the ILP solver backend that will be used.
Attempts Gurobi first; falls back to SCIP.