Alan R Lowe

The current `_run_tracker` function always calls `segmentation_to_objects` even if we are only changing some config parameters: https://github.com/lowe-lab-ucl/napari-btrack/blob/c294d1ca5c0c894599e464f9183464a2bcd8e4ff/napari_btrack/main.py#L144-L156 we could consider storing these detections, unless the chosen segmentation layer changes, in...

If an object is moving, it may make more sense to calculate the split geometry from the predicted position, rather than the last known position

segmentation field should use either `Image` or `Labels` layers as input

2

Currently, in quantumjot/napari-btrack#26 only `Image` layers can be used. However, it's common to use `Labels` layers for this purpose too, which has the advantage that they can be edited directly...

use example dataset to test plugin

1

https://rdr.ucl.ac.uk/articles/dataset/Cell_tracking_reference_dataset/16595978?file=31145701

@laureho

We already closed #64, but I thought I'd open a new issue to lower the bar to creating scivision compatible repos: https://cookiecutter.readthedocs.io/en/stable/

We should consider using a KNN to construct the intial graph, and compare this with the current triangulation method. https://pytorch-geometric.readthedocs.io/en/latest/generated/torch_geometric.nn.pool.knn_graph.html#torch_geometric.nn.pool.knn_graph This would have the additional benefit of allowing alternative hypotheses...

This will help with #36