currently, we are doing one neuron a time. the advantages are:
less boundary effect
avoid stitching across chunks, which could lead to errors
the disadvantages are:
the IO is redundant since we only use a very sparse part of each chunk. the IO time is longer than computation.
getting the skeletons around the neuron, we might be able to use skeleton to do agglomeration. the AI need to answer a question, whether this piece should merge to this neuron or not.
currently, we are doing one neuron a time. the advantages are:
the disadvantages are:
getting the skeletons around the neuron, we might be able to use skeleton to do agglomeration. the AI need to answer a question, whether this piece should merge to this neuron or not.
this has a very low priority.