Depends on brainglobe/cellfinder-core#220, with Keras 3.0, we should be able to use different DL backends. This could theoretically achieve multiple things:
Speed up (TF may not be the fastest)
Ease deployment (TF is not compatible with GPU on Windows, and isn't on conda-forge)
Depends on brainglobe/cellfinder-core#220, with Keras 3.0, we should be able to use different DL backends. This could theoretically achieve multiple things: