Why do you even need registerInput and registerOutput in IUffParser?

Why do we need to registerInput and registerOutput when the conversion process from frozen tf models to .uff and the uff parser should be able to easily parse the nodes from the graph directly?

Specifically, while there is some reason in being able to specify the input order in the case of convolutions (but then again internally it’s always converted to use NHWC anyway from what I understand), why do we even need to specify the input shapes?