Currently, the gala solver is extremely slow, but the proofreader is fast after an initial warmup. The gala solver can be sped up dramatically by abstracting out all of the node merge logic, and implementing merges only "virtually", using viridis (the merge tree).
The proofreader initial warmup can also be sped up by using a generic RAG as provided by scikit-image (see here for yet-to-be-implemented fast version.)
The solver/client relationship conforms to the interface defined in BigCat: https://github.com/saalfeldlab/bigcat/wiki/Actors,-responsibilities,-and-inter-process-communication
Currently, the gala solver is extremely slow, but the proofreader is fast after an initial warmup. The gala solver can be sped up dramatically by abstracting out all of the node merge logic, and implementing merges only "virtually", using viridis (the merge tree).
The proofreader initial warmup can also be sped up by using a generic RAG as provided by scikit-image (see here for yet-to-be-implemented fast version.)