instance segmentation pipeline - graph post processing

[stmartineau99]

graph postprocessing (postprocessing.py)

• draw_instances function - rasterize the instances or skeleton graph for easy viewing in napari
• clean_graph has 4 steps

1. remove_ticks reimplemented function from kimimaro, prune ticks (spurs) below a certain tick_length
2. split 3 degree junctions - separate the odd arm when it diverges from the through pair by at least min_branch_angle.
3. split 4 degree junctions, separating its two through pairs when they are collinear to within min_through_angle.
4. join close components at their endpoints - inspired by similar postporocessing approach in TARDIS library, search all endpoints for nearby endpoints that can be joined if they are within the join_dist and if they are sufficiently colinear, as determined by min_join_angle

tests/skeleton/test_postprocessing.py

• tests all stages of clean_graph

examples/skeleton/instance_segmentation.py

• script with CLI for the whole pipeline
  • starting from a binary mask, run teasar to get the skeleton, then convert to a graph and postprocess with clean_graph
  • convert to labels with connected_components - input to draw_instances
  • option to --view, open teasar skeleton and instances
• importantly, increase direction_span to get better tangent measurements, I found this helps best at around 8-10 to minimize aliasing artifacts