recursive two stage training - seung-lab/znn-release GitHub Wiki

Since the first stage output volume is smaller than the original channel and labled volume, we need to set offset parameter to align the output volume.

the offset value should be floor(FoV/2).