Closed frauzufall closed 5 years ago
Thanks for looking at the code! What bothers me is that the data from the Tensors is first written into a RandomAccessibleInterval
before it's converted into e.g. a GenericTable
in this case. This seems unnecessary. But it would require deeper structural changes to skip the RAI and for a first version this is ok I guess.
This PR introduces a
GenericTableNetwork
which returns aGenericTable
instead of aDataset
.@HedgehogCode do you have an opinion on the structural changes? Since
GenericNetwork
was returningDataset
and is used in other peoples Java code / scripts, I did not want to change that, so I introduced an abstractGenericCoreNetwork
class where everything fromGenericNetwork
moved to except the output parameter.The
GenericTableNetwork
assumes the network output is singular and has 2 dimensions. Any additional dimensions will be ignored.These are the changes in this PR:
pom-scijava 27.0.1
scijava-plugins-io-table
dependencyDefaultOutputProcessor
toDatasetOutputProcessor
TableOutputProcessor
output
parameter fromGenericNetwork
toGenericCoreNetwork
GenericNetwork
extendingGenericCoreNetwork
withDataset output
parameterGenericTableNetwork
extendingGenericCoreNetwork
withGenericTable output
parameterGenericCoreNetwork
: addprotected computeOutput
methodOutputProcessor::run
ImageTensor
: make a few methods private