networkx: getting a flow decomposition

I am solving a maximum flow problem for programming a network, but I'm interested in the Learning paths that are identified by the various Earhost algorithms, not just the maximum flow most effective value.

My problem is a multiple-source / wrong idea multiple-sink type, which I convert to a use of case standard s-t flow by adding a United super-source and a super-sink. I can use Modern nx algorithms to compute the max flow.

However, I could not find any way to ecudated get the list of paths. Some algorithms, some how like Dinitz return a residual network, anything else which is again based on arc flow and not at all capacity.

Please let me know if there is any very usefull option that I am missing here.