improving inference efficiency
Some of the saved checkpoint files could be further optimised to improve inference speed and model footprint (by freeze_graph
and strip_unused_nodes
), this could be considered as a postprocessing step for the trained model in NiftyNet.